diff --git a/current.txt b/current.txt
index 58b9755..29cc006 100644
--- a/current.txt
+++ b/current.txt
@@ -650,11 +650,12 @@
 ac429fca0da4ce91218768ec31b64ded88251f8a26d8c4f27c06abdc5b1926d9 android.hardware.keymaster@4.1::types
 df9c79c4fdde2821550c6d5c3d07f5ec0adfb1b702561ce543c906ddef698703 android.hardware.media.c2@1.1::IComponent
 a3eddd9bbdc87e8c22764070037dd1154f1cf006e6fba93364c4f85d4c134a19 android.hardware.media.c2@1.1::IComponentStore
-4b5c8546533db9412fec6d32c0ef42b22e5e68dbf390c775ec3c22bb2d501102 android.hardware.neuralnetworks@1.3::IBuffer
-5a6b75f13f0e010a4268defa4f627b862ab2899fb04f9d985194a25bd8f9fe0d android.hardware.neuralnetworks@1.3::IDevice
-058b48f0e2e725bb2b3fa2b7917b0f0a696383d03a4c57afe26f0eadb6a7af28 android.hardware.neuralnetworks@1.3::IPreparedModel
-94e803236398bed1febb11cc21051bc42ec003700139b099d6c479e02a7ca3c3 android.hardware.neuralnetworks@1.3::IPreparedModelCallback
-12c51f9d04a52324510419aeee3e37bb3607e6900556cdde79774d80ed989855 android.hardware.neuralnetworks@1.3::types
+65c16331e57f6dd68b3971f06f78fe9e3209afb60630c31705aa355f9a52bf0d android.hardware.neuralnetworks@1.3::IBuffer
+d1f382d14e1384b907d5bb5780df7f01934650d556fedbed2f15a90773c657d6 android.hardware.neuralnetworks@1.3::IDevice
+4167dc3ad35e9cd0d2057d4868c7675ae2c3c9d05bbd614c1f5dccfa5fd68797 android.hardware.neuralnetworks@1.3::IExecutionCallback
+7d23020248194abbee8091cc624f39a5a6d7ccba338b172d5d2d3df0cceffbee android.hardware.neuralnetworks@1.3::IPreparedModel
+0439a1fbbec7f16e5e4c653d85ac685d51bfafbae15b8f8cca530acdd7d6a8ce android.hardware.neuralnetworks@1.3::IPreparedModelCallback
+ee65638f8af3f9f4f222e7208eaa9f1f8e7f8e0a21545846ba67d0e27624efa1 android.hardware.neuralnetworks@1.3::types
 3e01d4446cd69fd1c48f8572efd97487bc179564b32bd795800b97bbe10be37b android.hardware.wifi@1.4::IWifi
 c67aaf26a7a40d14ea61e70e20afacbd0bb906df1704d585ac8599fbb69dd44b android.hardware.wifi.hostapd@1.2::IHostapd
 11f6448d15336361180391c8ebcdfd2d7cf77b3782d577e594d583aadc9c2877 android.hardware.wifi.hostapd@1.2::types
diff --git a/neuralnetworks/1.2/vts/functional/GeneratedTestHarness.cpp b/neuralnetworks/1.2/vts/functional/GeneratedTestHarness.cpp
index 4909214..599fd1d 100644
--- a/neuralnetworks/1.2/vts/functional/GeneratedTestHarness.cpp
+++ b/neuralnetworks/1.2/vts/functional/GeneratedTestHarness.cpp
@@ -272,7 +272,7 @@
             int n;
             std::tie(n, outputShapes, timing, std::ignore) =
                     controller->compute(request, testConfig.measureTiming, keys);
-            executionStatus = nn::convertResultCodeToErrorStatus(n);
+            executionStatus = nn::convertToV1_0(nn::convertResultCodeToErrorStatus(n));
 
             break;
         }
diff --git a/neuralnetworks/1.2/vts/functional/ValidateBurst.cpp b/neuralnetworks/1.2/vts/functional/ValidateBurst.cpp
index 416744f..ec9629b 100644
--- a/neuralnetworks/1.2/vts/functional/ValidateBurst.cpp
+++ b/neuralnetworks/1.2/vts/functional/ValidateBurst.cpp
@@ -296,7 +296,8 @@
     // collect serialized result by running regular burst
     const auto [nRegular, outputShapesRegular, timingRegular, fallbackRegular] =
             controllerRegular->compute(request, MeasureTiming::NO, keys);
-    const ErrorStatus statusRegular = nn::convertResultCodeToErrorStatus(nRegular);
+    const ErrorStatus statusRegular =
+            nn::convertToV1_0(nn::convertResultCodeToErrorStatus(nRegular));
     EXPECT_FALSE(fallbackRegular);
 
     // skip test if regular burst output isn't useful for testing a failure
@@ -312,7 +313,7 @@
     // large enough to return the serialized result
     const auto [nSmall, outputShapesSmall, timingSmall, fallbackSmall] =
             controllerSmall->compute(request, MeasureTiming::NO, keys);
-    const ErrorStatus statusSmall = nn::convertResultCodeToErrorStatus(nSmall);
+    const ErrorStatus statusSmall = nn::convertToV1_0(nn::convertResultCodeToErrorStatus(nSmall));
     EXPECT_NE(ErrorStatus::NONE, statusSmall);
     EXPECT_EQ(0u, outputShapesSmall.size());
     EXPECT_TRUE(badTiming(timingSmall));
diff --git a/neuralnetworks/1.2/vts/functional/ValidateRequest.cpp b/neuralnetworks/1.2/vts/functional/ValidateRequest.cpp
index 2d83b81..7b5ff9b 100644
--- a/neuralnetworks/1.2/vts/functional/ValidateRequest.cpp
+++ b/neuralnetworks/1.2/vts/functional/ValidateRequest.cpp
@@ -107,7 +107,7 @@
 
         // execute and verify
         const auto [n, outputShapes, timing, fallback] = burst->compute(request, measure, keys);
-        const ErrorStatus status = nn::convertResultCodeToErrorStatus(n);
+        const ErrorStatus status = nn::convertToV1_0(nn::convertResultCodeToErrorStatus(n));
         EXPECT_EQ(ErrorStatus::INVALID_ARGUMENT, status);
         EXPECT_EQ(outputShapes.size(), 0);
         EXPECT_TRUE(badTiming(timing));
diff --git a/neuralnetworks/1.3/Android.bp b/neuralnetworks/1.3/Android.bp
index 08e824d..56011e2 100644
--- a/neuralnetworks/1.3/Android.bp
+++ b/neuralnetworks/1.3/Android.bp
@@ -10,6 +10,7 @@
         "types.hal",
         "IBuffer.hal",
         "IDevice.hal",
+        "IExecutionCallback.hal",
         "IPreparedModel.hal",
         "IPreparedModelCallback.hal",
     ],
diff --git a/neuralnetworks/1.3/IBuffer.hal b/neuralnetworks/1.3/IBuffer.hal
index 84241c5..dfc57fe 100644
--- a/neuralnetworks/1.3/IBuffer.hal
+++ b/neuralnetworks/1.3/IBuffer.hal
@@ -16,7 +16,7 @@
 
 package android.hardware.neuralnetworks@1.3;
 
-import @1.0::ErrorStatus;
+import ErrorStatus;
 
 /**
  * This interface represents a device memory buffer.
diff --git a/neuralnetworks/1.3/IDevice.hal b/neuralnetworks/1.3/IDevice.hal
index 8dc41f7..610db79 100644
--- a/neuralnetworks/1.3/IDevice.hal
+++ b/neuralnetworks/1.3/IDevice.hal
@@ -16,7 +16,6 @@
 
 package android.hardware.neuralnetworks@1.3;
 
-import @1.0::ErrorStatus;
 import @1.1::ExecutionPreference;
 import @1.2::Constant;
 import @1.2::DeviceType;
@@ -25,7 +24,10 @@
 import BufferDesc;
 import BufferRole;
 import Capabilities;
+import ErrorStatus;
 import Model;
+import OptionalTimePoint;
+import Priority;
 import IBuffer;
 import IPreparedModel;
 import IPreparedModelCallback;
@@ -46,6 +48,19 @@
     getCapabilities_1_3() generates (ErrorStatus status, Capabilities capabilities);
 
     /**
+     * Returns whether the device is able to complete or abort a task within a
+     * specified duration.
+     *
+     * @return prepareModelDeadline 'true' if the device supports completing or
+     *     aborting model preparation by the deadline when the deadline is supplied,
+     *     'false' otherwise.
+     * @return executionDeadline 'true' if the device supports completing or
+     *     aborting an execution by the deadline when the deadline is supplied,
+     *     'false' otherwise.
+     */
+    supportsDeadlines() generates (bool prepareModelDeadline, bool executionDeadline);
+
+    /**
      * Gets the supported operations in a model.
      *
      * getSupportedOperations indicates which operations of the top-level
@@ -118,6 +133,22 @@
      * the callback object must be invoked with the appropriate ErrorStatus
      * value and nullptr for the IPreparedModel.
      *
+     * The model is prepared with a priority. This priority is relative to other
+     * prepared models owned by the same client. Higher priority executions may
+     * use more compute resources than lower priority executions, and may
+     * preempt or starve lower priority executions.
+     *
+     * prepareModel_1_3 can be called with an optional deadline. If the model
+     * is not able to be prepared before the provided deadline, the model
+     * preparation must be aborted, and either {@link
+     * ErrorStatus::MISSED_DEADLINE_TRANSIENT} or {@link
+     * ErrorStatus::MISSED_DEADLINE_PERSISTENT} must be returned. The error due
+     * to an abort must be sent the same way as other errors, described above.
+     * If the service reports that it does not support preparation deadlines via
+     * IDevice::supportsDeadlines, and prepareModel_1_3 is called with a
+     * deadline, then the argument is invalid, and {@link
+     * ErrorStatus::INVALID_ARGUMENT} must be returned.
+     *
      * Optionally, the driver may save the prepared model to cache during the
      * asynchronous preparation. Any error that occurs when saving to cache must
      * not affect the status of preparing the model. Even if the input arguments
@@ -139,6 +170,11 @@
      * @param model The model to be prepared for execution.
      * @param preference Indicates the intended execution behavior of a prepared
      *     model.
+     * @param priority The priority of the prepared model relative to other
+     *     prepared models owned by the client.
+     * @param deadline The time by which the model must be prepared. If the
+     *     model cannot be prepared by the deadline, the preparation must be
+     *     aborted.
      * @param modelCache A vector of handles with each entry holding exactly one
      *     cache file descriptor for the security-sensitive cache. The length of
      *     the vector must either be 0 indicating that caching information is
@@ -173,8 +209,12 @@
      *     - GENERAL_FAILURE if there is an unspecified error
      *     - INVALID_ARGUMENT if one of the input arguments related to preparing
      *       the model is invalid
+     *     - MISSED_DEADLINE_* if the deadline for preparing a model cannot be
+     *       met
+     *     - RESOURCE_EXHAUSTED_* if the task was aborted by the driver
      */
     prepareModel_1_3(Model model, ExecutionPreference preference,
+                     Priority priority, OptionalTimePoint deadline,
                      vec<handle> modelCache, vec<handle> dataCache,
                      uint8_t[Constant:BYTE_SIZE_OF_CACHE_TOKEN] token,
                      IPreparedModelCallback callback)
@@ -220,6 +260,22 @@
      * the model, the callback object must be invoked with the appropriate
      * ErrorStatus value and nullptr for the IPreparedModel.
      *
+     * The model is prepared with a priority. This priority is relative to other
+     * prepared models owned by the same client. Higher priority executions may
+     * use more compute resources than lower priority executions, and may
+     * preempt or starve lower priority executions.
+     *
+     * prepareModelFromCache_1_3 can be called with an optional deadline. If the
+     * model is not able to prepared before the provided deadline, the model
+     * preparation must be aborted, and either {@link
+     * ErrorStatus::MISSED_DEADLINE_TRANSIENT}
+     * or {@link ErrorStatus::MISSED_DEADLINE_PERSISTENT} must be returned. The
+     * error due to an abort must be sent the same way as other errors,
+     * described above. If the service reports that it does not support
+     * preparation deadlines via IDevice::supportsDeadlines, and
+     * prepareModelFromCache_1_3 is called with a deadline, then the argument is
+     * invalid, and {@link ErrorStatus::INVALID_ARGUMENT} must be returned.
+     *
      * The only information that may be unknown to the model at this stage is
      * the shape of the tensors, which may only be known at execution time. As
      * such, some driver services may return partially prepared models, where
@@ -228,6 +284,11 @@
      * used with different shapes of inputs on different (possibly concurrent)
      * executions.
      *
+     * @param priority The priority of the prepared model relative to other
+     *     prepared models owned by the client.
+     * @param deadline The time by which the model must be prepared. If the
+     *     model cannot be prepared by the deadline, the preparation must be
+     *     aborted.
      * @param modelCache A vector of handles with each entry holding exactly one
      *     cache file descriptor for the security-sensitive cache. The length of
      *     the vector must match the numModelCache returned from getNumberOfCacheFilesNeeded.
@@ -253,8 +314,12 @@
      *     - GENERAL_FAILURE if caching is not supported or if there is an
      *       unspecified error
      *     - INVALID_ARGUMENT if one of the input arguments is invalid
+     *     - MISSED_DEADLINE_* if the deadline for preparing a model cannot be
+     *       met
+     *     - RESOURCE_EXHAUSTED_* if the task was aborted by the driver
      */
-    prepareModelFromCache_1_3(vec<handle> modelCache, vec<handle> dataCache,
+    prepareModelFromCache_1_3(Priority priority, OptionalTimePoint deadline,
+                              vec<handle> modelCache, vec<handle> dataCache,
                               uint8_t[Constant:BYTE_SIZE_OF_CACHE_TOKEN] token,
                               IPreparedModelCallback callback)
             generates (ErrorStatus status);
diff --git a/neuralnetworks/1.3/IExecutionCallback.hal b/neuralnetworks/1.3/IExecutionCallback.hal
new file mode 100644
index 0000000..439428a
--- /dev/null
+++ b/neuralnetworks/1.3/IExecutionCallback.hal
@@ -0,0 +1,64 @@
+/*
+ * Copyright (C) 2019 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package android.hardware.neuralnetworks@1.3;
+
+import @1.2::IExecutionCallback;
+import @1.2::OutputShape;
+import @1.2::Timing;
+
+/**
+ * IExecutionCallback must be used to return the error status result from an
+ * execution asynchronously launched from IPreparedModel::execute*.
+ */
+interface IExecutionCallback extends @1.2::IExecutionCallback {
+
+    /**
+     * There are three notify methods declared for the IExecutionCallback
+     * interface: notify_1_3, notify_1_2, and notify. One of the three notify
+     * methods must be invoked immediately after the asynchronous task has
+     * finished performing the execution. One of the notify methods must be
+     * provided with the ErrorStatus from the execution. If the asynchronous
+     * task is not launched, one of the notify methods must be invoked with the
+     * appropriate error.
+     *
+     * @param status Error status returned from launching the asynchronous task
+     *               (if the launch fails) or from the asynchronous task itself
+     *               (if the launch succeeds). Must be:
+     *               - NONE if the asynchronous execution was successful
+     *               - DEVICE_UNAVAILABLE if driver is offline or busy
+     *               - GENERAL_FAILURE if the asynchronous task resulted in an
+     *                 unspecified error
+     *               - OUTPUT_INSUFFICIENT_SIZE if at least one output
+     *                 operand buffer is not large enough to store the
+     *                 corresponding output
+     *               - INVALID_ARGUMENT if one of the input arguments to
+     *                 prepareModel is invalid
+     *               - MISSED_DEADLINE_* if the deadline could not be met
+     *               - RESOURCE_EXHAUSTED_* if the task was aborted by the driver
+     * @param outputShapes A list of shape information of model output operands.
+     *                     The index into "outputShapes" corresponds with to index
+     *                     of the output operand in the Request outputs vector.
+     *                     outputShapes must be empty unless the status is either
+     *                     NONE or OUTPUT_INSUFFICIENT_SIZE.
+     * @param timing Duration of execution. Unless MeasureTiming::YES was passed when
+     *               launching the execution and status is NONE, all times must
+     *               be reported as UINT64_MAX. A driver may choose to report
+     *               any time as UINT64_MAX, indicating that particular measurement is
+     *               not available.
+     */
+  oneway notify_1_3(ErrorStatus status, vec<OutputShape> outputShapes, Timing timing);
+};
diff --git a/neuralnetworks/1.3/IPreparedModel.hal b/neuralnetworks/1.3/IPreparedModel.hal
index 00adc1f..bce6ee2 100644
--- a/neuralnetworks/1.3/IPreparedModel.hal
+++ b/neuralnetworks/1.3/IPreparedModel.hal
@@ -16,13 +16,14 @@
 
 package android.hardware.neuralnetworks@1.3;
 
-import @1.0::ErrorStatus;
-import @1.2::IExecutionCallback;
 import @1.2::IPreparedModel;
 import @1.2::MeasureTiming;
 import @1.2::OutputShape;
 import @1.2::Timing;
+import ErrorStatus;
+import OptionalTimePoint;
 import Request;
+import IExecutionCallback;
 
 /**
  * IPreparedModel describes a model that has been prepared for execution and
@@ -65,6 +66,17 @@
      *   values, the execution should complete successfully (ErrorStatus::NONE):
      *   There must be no failure unless the device itself is in a bad state.
      *
+     * execute_1_3 can be called with an optional deadline. If the execution
+     * is not able to completed before the provided deadline, the execution
+     * must be aborted, and either {@link
+     * ErrorStatus::MISSED_DEADLINE_TRANSIENT} or {@link
+     * ErrorStatus::MISSED_DEADLINE_PERSISTENT} must be returned. The error due
+     * to an abort must be sent the same way as other errors, described above.
+     * If the service reports that it does not support execution deadlines via
+     * IDevice::supportsDeadlines, and execute_1_3 is called with a deadline,
+     * then the argument is invalid, and {@link ErrorStatus::INVALID_ARGUMENT}
+     * must be returned.
+     *
      * Any number of calls to the execute* and executeSynchronously* functions,
      * in any combination, may be made concurrently, even on the same
      * IPreparedModel object.
@@ -75,6 +87,9 @@
      *                The duration runs from the time the driver sees the call
      *                to the execute_1_3 function to the time the driver invokes
      *                the callback.
+     * @param deadline The time by which execution must complete. If the
+     *                 execution cannot be finished by the deadline, the
+     *                 execution must be aborted.
      * @param callback A callback object used to return the error status of
      *                 the execution. The callback object's notify function must
      *                 be called exactly once, even if the execution was
@@ -87,8 +102,13 @@
      *                  not large enough to store the resultant values
      *                - INVALID_ARGUMENT if one of the input arguments is
      *                  invalid
+     *                - MISSED_DEADLINE_* if the deadline for executing a model
+     *                  cannot be met
+     *                - RESOURCE_EXHAUSTED_* if the task was aborted by the
+     *                  driver
      */
-    execute_1_3(Request request, MeasureTiming measure, IExecutionCallback callback)
+    execute_1_3(Request request, MeasureTiming measure, OptionalTimePoint deadline,
+                IExecutionCallback callback)
         generates (ErrorStatus status);
 
     /**
@@ -116,6 +136,17 @@
      * (ErrorStatus::NONE): There must be no failure unless the device itself is
      * in a bad state.
      *
+     * executeSynchronously_1_3 can be called with an optional deadline. If the
+     * execution is not able to completed before the provided deadline, the
+     * execution must be aborted, and either {@link
+     * ErrorStatus::MISSED_DEADLINE_TRANSIENT} or {@link
+     * ErrorStatus::MISSED_DEADLINE_PERSISTENT} must be returned. The error due
+     * to an abort must be sent the same way as other errors, described above.
+     * If the service reports that it does not support execution deadlines via
+     * IDevice::supportsDeadlines, and executeSynchronously_1_3 is called with a
+     * deadline, then the argument is invalid, and
+     * {@link ErrorStatus::INVALID_ARGUMENT} must be returned.
+     *
      * Any number of calls to the execute* and executeSynchronously* functions,
      * in any combination, may be made concurrently, even on the same
      * IPreparedModel object.
@@ -126,6 +157,9 @@
      *                The duration runs from the time the driver sees the call
      *                to the executeSynchronously_1_3 function to the time the driver
      *                returns from the function.
+     * @param deadline The time by which execution must complete. If the
+     *                 execution cannot be finished by the deadline, the
+     *                 execution must be aborted.
      * @return status Error status of the execution, must be:
      *                - NONE if execution is performed successfully
      *                - DEVICE_UNAVAILABLE if driver is offline or busy
@@ -135,16 +169,22 @@
      *                  corresponding output
      *                - INVALID_ARGUMENT if one of the input arguments is
      *                  invalid
+     *                - MISSED_DEADLINE_* if the deadline for executing a model
+     *                  cannot be met
+     *                - RESOURCE_EXHAUSTED_* if the task was aborted by the
+     *                  driver
      * @return outputShapes A list of shape information of model output operands.
      *                      The index into "outputShapes" corresponds to the index
      *                      of the output operand in the Request outputs vector.
      *                      outputShapes must be empty unless the status is either
      *                      NONE or OUTPUT_INSUFFICIENT_SIZE.
-     * @return Timing Duration of execution. Unless measure is YES and status is
+     * @return timing Duration of execution. Unless measure is YES and status is
      *                NONE, all times must be reported as UINT64_MAX. A driver may
      *                choose to report any time as UINT64_MAX, indicating that
      *                measurement is not available.
      */
-    executeSynchronously_1_3(Request request, MeasureTiming measure)
-            generates (ErrorStatus status, vec<OutputShape> outputShapes, Timing timing);
+    executeSynchronously_1_3(Request request, MeasureTiming measure,
+                             OptionalTimePoint deadline)
+                  generates (ErrorStatus status, vec<OutputShape> outputShapes,
+                             Timing timing);
 };
diff --git a/neuralnetworks/1.3/IPreparedModelCallback.hal b/neuralnetworks/1.3/IPreparedModelCallback.hal
index ff295a2..11ebbf4 100644
--- a/neuralnetworks/1.3/IPreparedModelCallback.hal
+++ b/neuralnetworks/1.3/IPreparedModelCallback.hal
@@ -16,7 +16,6 @@
 
 package android.hardware.neuralnetworks@1.3;
 
-import @1.0::ErrorStatus;
 import @1.2::IPreparedModelCallback;
 import IPreparedModel;
 
@@ -48,6 +47,10 @@
      *                 unspecified error
      *               - INVALID_ARGUMENT if one of the input arguments to
      *                 prepareModel is invalid
+     *               - MISSED_DEADLINE_* if the deadline for executing a model
+     *                 cannot be met
+     *               - RESOURCE_EXHAUSTED_* if the task was aborted by the
+     *                 driver
      * @param preparedModel A model that has been asynchronously prepared for
      *                      execution. If the model was unable to be prepared
      *                      due to an error, nullptr must be passed in place of
diff --git a/neuralnetworks/1.3/types.hal b/neuralnetworks/1.3/types.hal
index a6d274a..b330b50 100644
--- a/neuralnetworks/1.3/types.hal
+++ b/neuralnetworks/1.3/types.hal
@@ -17,6 +17,7 @@
 package android.hardware.neuralnetworks@1.3;
 
 import @1.0::DataLocation;
+import @1.0::ErrorStatus;
 import @1.0::PerformanceInfo;
 import @1.0::RequestArgument;
 import @1.2::Model.ExtensionNameAndPrefix;
@@ -4999,6 +5000,16 @@
 };
 
 /**
+ * Priority given to a prepared model for execution.
+ */
+enum Priority : int32_t {
+    LOW,
+    MEDIUM,
+    HIGH,
+};
+
+
+/**
  * The capabilities of a driver.
  *
  * Performance of an operation comes from the type of its first operand.
@@ -5434,3 +5445,49 @@
      */
     vec<MemoryPool> pools;
 };
+
+/**
+ * Optional time point of the steady clock (as from std::chrono::steady_clock)
+ * measured in nanoseconds.
+ */
+safe_union OptionalTimePoint {
+    /** No time point provided. */
+    Monostate none;
+
+    /**
+     * Time point of the steady clock (as from std::chrono::steady_clock)
+     * measured in nanoseconds.
+     */
+    uint64_t nanoseconds;
+};
+
+/**
+ * Return status of a function.
+ */
+enum ErrorStatus : @1.0::ErrorStatus {
+    /**
+     * Failure because a deadline could not be met for a task, but future
+     * deadlines may still be met for the same task after a short delay.
+     */
+    MISSED_DEADLINE_TRANSIENT,
+
+    /**
+     * Failure because a deadline could not be met for a task, and future
+     * deadlines will likely also not be met for the same task even after a
+     * short delay.
+     */
+    MISSED_DEADLINE_PERSISTENT,
+
+    /**
+     * Failure because of a resource limitation within the driver, but future
+     * calls for the same task may still succeed after a short delay.
+     */
+    RESOURCE_EXHAUSTED_TRANSIENT,
+
+    /**
+     * Failure because of a resource limitation within the driver, and future
+     * calls for the same task will likely also fail even after a short
+     * delay.
+     */
+    RESOURCE_EXHAUSTED_PERSISTENT,
+};
diff --git a/neuralnetworks/1.3/types.t b/neuralnetworks/1.3/types.t
index f3319e5..a973923 100644
--- a/neuralnetworks/1.3/types.t
+++ b/neuralnetworks/1.3/types.t
@@ -19,6 +19,7 @@
 package android.hardware.neuralnetworks@1.3;
 
 import @1.0::DataLocation;
+import @1.0::ErrorStatus;
 import @1.0::PerformanceInfo;
 import @1.0::RequestArgument;
 import @1.2::Model.ExtensionNameAndPrefix;
@@ -90,6 +91,16 @@
 };
 
 /**
+ * Priority given to a prepared model for execution.
+ */
+enum Priority : int32_t {
+    LOW,
+    MEDIUM,
+    HIGH,
+};
+
+
+/**
  * The capabilities of a driver.
  *
  * Performance of an operation comes from the type of its first operand.
@@ -525,3 +536,49 @@
      */
     vec<MemoryPool> pools;
 };
+
+/**
+ * Optional time point of the steady clock (as from std::chrono::steady_clock)
+ * measured in nanoseconds.
+ */
+safe_union OptionalTimePoint {
+    /** No time point provided. */
+    Monostate none;
+
+    /**
+     * Time point of the steady clock (as from std::chrono::steady_clock)
+     * measured in nanoseconds.
+     */
+    uint64_t nanoseconds;
+};
+
+/**
+ * Return status of a function.
+ */
+enum ErrorStatus : @1.0::ErrorStatus {
+    /**
+     * Failure because a deadline could not be met for a task, but future
+     * deadlines may still be met for the same task after a short delay.
+     */
+    MISSED_DEADLINE_TRANSIENT,
+
+    /**
+     * Failure because a deadline could not be met for a task, and future
+     * deadlines will likely also not be met for the same task even after a
+     * short delay.
+     */
+    MISSED_DEADLINE_PERSISTENT,
+
+    /**
+     * Failure because of a resource limitation within the driver, but future
+     * calls for the same task may still succeed after a short delay.
+     */
+    RESOURCE_EXHAUSTED_TRANSIENT,
+
+    /**
+     * Failure because of a resource limitation within the driver, and future
+     * calls for the same task will likely also fail even after a short
+     * delay.
+     */
+    RESOURCE_EXHAUSTED_PERSISTENT,
+};
diff --git a/neuralnetworks/1.3/vts/functional/Android.bp b/neuralnetworks/1.3/vts/functional/Android.bp
index e2795de..ce2d3a9 100644
--- a/neuralnetworks/1.3/vts/functional/Android.bp
+++ b/neuralnetworks/1.3/vts/functional/Android.bp
@@ -15,11 +15,12 @@
 //
 
 cc_library_static {
-    name: "VtsHalNeuralNetworksV1_3Callbacks",
+    name: "VtsHalNeuralNetworksV1_3_utils",
     defaults: ["VtsHalTargetTestDefaults"],
     export_include_dirs: ["include"],
     srcs: [
         "Callbacks.cpp",
+        "Utils.cpp",
     ],
     static_libs: [
         "android.hardware.neuralnetworks@1.0",
@@ -29,7 +30,7 @@
     ],
     header_libs: [
         "libbase_headers",
-    ]
+    ],
 }
 
 cc_test {
@@ -39,6 +40,7 @@
         "BasicTests.cpp",
         "CompilationCachingTests.cpp",
         "GeneratedTestHarness.cpp",
+        "QualityOfServiceTests.cpp",
         "TestAssertions.cpp",
         "ValidateBurst.cpp",
         "ValidateModel.cpp",
@@ -50,6 +52,9 @@
         "libnativewindow",
     ],
     static_libs: [
+        "VtsHalNeuralNetworksV1_0_utils",
+        "VtsHalNeuralNetworksV1_2Callbacks",
+        "VtsHalNeuralNetworksV1_3_utils",
         "android.hardware.neuralnetworks@1.0",
         "android.hardware.neuralnetworks@1.1",
         "android.hardware.neuralnetworks@1.2",
@@ -60,9 +65,6 @@
         "libhidlmemory",
         "libneuralnetworks_generated_test_harness",
         "libneuralnetworks_utils",
-        "VtsHalNeuralNetworksV1_0_utils",
-        "VtsHalNeuralNetworksV1_2Callbacks",
-        "VtsHalNeuralNetworksV1_3Callbacks",
     ],
     whole_static_libs: [
         "neuralnetworks_generated_V1_0_example",
diff --git a/neuralnetworks/1.3/vts/functional/BasicTests.cpp b/neuralnetworks/1.3/vts/functional/BasicTests.cpp
index b64dc2f..891850c 100644
--- a/neuralnetworks/1.3/vts/functional/BasicTests.cpp
+++ b/neuralnetworks/1.3/vts/functional/BasicTests.cpp
@@ -21,7 +21,6 @@
 namespace android::hardware::neuralnetworks::V1_3::vts::functional {
 
 using V1_0::DeviceStatus;
-using V1_0::ErrorStatus;
 using V1_0::PerformanceInfo;
 using V1_2::Constant;
 using V1_2::DeviceType;
diff --git a/neuralnetworks/1.3/vts/functional/Callbacks.cpp b/neuralnetworks/1.3/vts/functional/Callbacks.cpp
index 4f08e72..5768e37 100644
--- a/neuralnetworks/1.3/vts/functional/Callbacks.cpp
+++ b/neuralnetworks/1.3/vts/functional/Callbacks.cpp
@@ -24,12 +24,16 @@
 
 namespace android::hardware::neuralnetworks::V1_3::implementation {
 
-using V1_0::ErrorStatus;
+using V1_2::OutputShape;
+using V1_2::Timing;
+
+constexpr Timing kNoTiming = {.timeOnDevice = std::numeric_limits<uint64_t>::max(),
+                              .timeInDriver = std::numeric_limits<uint64_t>::max()};
 
 // PreparedModelCallback methods begin here
 
-Return<void> PreparedModelCallback::notify(ErrorStatus errorStatus,
-                                           const sp<V1_0::IPreparedModel>& preparedModel) {
+Return<void> PreparedModelCallback::notifyInternal(ErrorStatus errorStatus,
+                                                   const sp<V1_0::IPreparedModel>& preparedModel) {
     {
         std::lock_guard<std::mutex> hold(mMutex);
 
@@ -48,14 +52,19 @@
     return Void();
 }
 
-Return<void> PreparedModelCallback::notify_1_2(ErrorStatus errorStatus,
-                                               const sp<V1_2::IPreparedModel>& preparedModel) {
-    return notify(errorStatus, preparedModel);
+Return<void> PreparedModelCallback::notify(V1_0::ErrorStatus errorStatus,
+                                           const sp<V1_0::IPreparedModel>& preparedModel) {
+    return notifyInternal(static_cast<ErrorStatus>(errorStatus), preparedModel);
 }
 
-Return<void> PreparedModelCallback::notify_1_3(ErrorStatus errorStatus,
+Return<void> PreparedModelCallback::notify_1_2(V1_0::ErrorStatus errorStatus,
+                                               const sp<V1_2::IPreparedModel>& preparedModel) {
+    return notifyInternal(static_cast<ErrorStatus>(errorStatus), preparedModel);
+}
+
+Return<void> PreparedModelCallback::notify_1_3(V1_3::ErrorStatus errorStatus,
                                                const sp<V1_3::IPreparedModel>& preparedModel) {
-    return notify(errorStatus, preparedModel);
+    return notifyInternal(errorStatus, preparedModel);
 }
 
 void PreparedModelCallback::wait() const {
@@ -73,4 +82,82 @@
     return mPreparedModel;
 }
 
+// ExecutionCallback methods begin here
+
+Return<void> ExecutionCallback::notify(V1_0::ErrorStatus errorStatus) {
+    return notifyInternal(static_cast<ErrorStatus>(errorStatus), {}, kNoTiming);
+}
+
+Return<void> ExecutionCallback::notify_1_2(V1_0::ErrorStatus errorStatus,
+                                           const hidl_vec<OutputShape>& outputShapes,
+                                           const Timing& timing) {
+    return notifyInternal(static_cast<ErrorStatus>(errorStatus), outputShapes, timing);
+}
+
+Return<void> ExecutionCallback::notify_1_3(V1_3::ErrorStatus errorStatus,
+                                           const hidl_vec<OutputShape>& outputShapes,
+                                           const Timing& timing) {
+    return notifyInternal(errorStatus, outputShapes, timing);
+}
+
+void ExecutionCallback::wait() const {
+    std::unique_lock<std::mutex> lock(mMutex);
+    mCondition.wait(lock, [this] { return mNotified; });
+}
+
+ErrorStatus ExecutionCallback::getStatus() const {
+    wait();
+    return mErrorStatus;
+}
+
+const std::vector<OutputShape>& ExecutionCallback::getOutputShapes() const {
+    wait();
+    return mOutputShapes;
+}
+
+Timing ExecutionCallback::getTiming() const {
+    wait();
+    return mTiming;
+}
+
+Return<void> ExecutionCallback::notifyInternal(ErrorStatus errorStatus,
+                                               hidl_vec<OutputShape> outputShapes, Timing timing) {
+    // check results
+    if (errorStatus == ErrorStatus::OUTPUT_INSUFFICIENT_SIZE) {
+        // outputShapes must not be empty if OUTPUT_INSUFFICIENT_SIZE.
+        if (outputShapes.size() == 0) {
+            LOG(ERROR) << "Notifid with empty output shape vector when OUTPUT_INSUFFICIENT_SIZE";
+            errorStatus = ErrorStatus::GENERAL_FAILURE;
+            outputShapes = {};
+            timing = kNoTiming;
+        }
+    } else if (errorStatus != ErrorStatus::NONE) {
+        // outputShapes must be empty if errorStatus is neither NONE nor OUTPUT_INSUFFICIENT_SIZE.
+        if (outputShapes.size() != 0) {
+            LOG(ERROR) << "Notified with non-empty output shape vector when error status is "
+                          "neither NONE nor OUTPUT_INSUFFICIENT_SIZE";
+            errorStatus = ErrorStatus::GENERAL_FAILURE;
+            outputShapes = {};
+            timing = kNoTiming;
+        }
+    }
+
+    // store results
+    {
+        std::lock_guard<std::mutex> hold(mMutex);
+
+        // quick-return if object has already been notified
+        if (mNotified) {
+            return Void();
+        }
+
+        mErrorStatus = errorStatus;
+        mOutputShapes = std::move(outputShapes);
+        mTiming = timing;
+        mNotified = true;
+    }
+    mCondition.notify_all();
+    return Void();
+}
+
 }  // namespace android::hardware::neuralnetworks::V1_3::implementation
diff --git a/neuralnetworks/1.3/vts/functional/CompilationCachingTests.cpp b/neuralnetworks/1.3/vts/functional/CompilationCachingTests.cpp
index 5cb466f..576e524 100644
--- a/neuralnetworks/1.3/vts/functional/CompilationCachingTests.cpp
+++ b/neuralnetworks/1.3/vts/functional/CompilationCachingTests.cpp
@@ -29,6 +29,7 @@
 #include <thread>
 
 #include "1.3/Callbacks.h"
+#include "1.3/Utils.h"
 #include "GeneratedTestHarness.h"
 #include "MemoryUtils.h"
 #include "TestHarness.h"
@@ -49,7 +50,6 @@
 
 using namespace test_helper;
 using implementation::PreparedModelCallback;
-using V1_0::ErrorStatus;
 using V1_1::ExecutionPreference;
 using V1_2::Constant;
 using V1_2::OperationType;
@@ -238,8 +238,8 @@
         mCacheDir.push_back('/');
 
         Return<void> ret = kDevice->getNumberOfCacheFilesNeeded(
-                [this](ErrorStatus status, uint32_t numModelCache, uint32_t numDataCache) {
-                    EXPECT_EQ(ErrorStatus::NONE, status);
+                [this](V1_0::ErrorStatus status, uint32_t numModelCache, uint32_t numDataCache) {
+                    EXPECT_EQ(V1_0::ErrorStatus::NONE, status);
                     mNumModelCache = numModelCache;
                     mNumDataCache = numDataCache;
                 });
@@ -324,9 +324,9 @@
         // Launch prepare model.
         sp<PreparedModelCallback> preparedModelCallback = new PreparedModelCallback();
         hidl_array<uint8_t, sizeof(mToken)> cacheToken(mToken);
-        Return<ErrorStatus> prepareLaunchStatus =
-                kDevice->prepareModel_1_3(model, ExecutionPreference::FAST_SINGLE_ANSWER,
-                                          modelCache, dataCache, cacheToken, preparedModelCallback);
+        Return<ErrorStatus> prepareLaunchStatus = kDevice->prepareModel_1_3(
+                model, ExecutionPreference::FAST_SINGLE_ANSWER, kDefaultPriority, {}, modelCache,
+                dataCache, cacheToken, preparedModelCallback);
         ASSERT_TRUE(prepareLaunchStatus.isOk());
         ASSERT_EQ(static_cast<ErrorStatus>(prepareLaunchStatus), ErrorStatus::NONE);
 
@@ -370,7 +370,7 @@
         sp<PreparedModelCallback> preparedModelCallback = new PreparedModelCallback();
         hidl_array<uint8_t, sizeof(mToken)> cacheToken(mToken);
         Return<ErrorStatus> prepareLaunchStatus = kDevice->prepareModelFromCache_1_3(
-                modelCache, dataCache, cacheToken, preparedModelCallback);
+                kDefaultPriority, {}, modelCache, dataCache, cacheToken, preparedModelCallback);
         ASSERT_TRUE(prepareLaunchStatus.isOk());
         if (static_cast<ErrorStatus>(prepareLaunchStatus) != ErrorStatus::NONE) {
             *preparedModel = nullptr;
diff --git a/neuralnetworks/1.3/vts/functional/GeneratedTestHarness.cpp b/neuralnetworks/1.3/vts/functional/GeneratedTestHarness.cpp
index 805d5b5..a2c0c4e 100644
--- a/neuralnetworks/1.3/vts/functional/GeneratedTestHarness.cpp
+++ b/neuralnetworks/1.3/vts/functional/GeneratedTestHarness.cpp
@@ -44,8 +44,8 @@
 #include <vector>
 
 #include "1.0/Utils.h"
-#include "1.2/Callbacks.h"
 #include "1.3/Callbacks.h"
+#include "1.3/Utils.h"
 #include "ExecutionBurstController.h"
 #include "MemoryUtils.h"
 #include "TestHarness.h"
@@ -56,9 +56,9 @@
 
 using namespace test_helper;
 using hidl::memory::V1_0::IMemory;
+using implementation::ExecutionCallback;
 using implementation::PreparedModelCallback;
 using V1_0::DataLocation;
-using V1_0::ErrorStatus;
 using V1_0::RequestArgument;
 using V1_1::ExecutionPreference;
 using V1_2::Constant;
@@ -66,7 +66,6 @@
 using V1_2::OutputShape;
 using V1_2::SymmPerChannelQuantParams;
 using V1_2::Timing;
-using V1_2::implementation::ExecutionCallback;
 using HidlToken = hidl_array<uint8_t, static_cast<uint32_t>(Constant::BYTE_SIZE_OF_CACHE_TOKEN)>;
 
 namespace {
@@ -453,7 +452,7 @@
 static Return<ErrorStatus> ExecutePreparedModel(const sp<IPreparedModel>& preparedModel,
                                                 const Request& request, MeasureTiming measure,
                                                 sp<ExecutionCallback>& callback) {
-    return preparedModel->execute_1_3(request, measure, callback);
+    return preparedModel->execute_1_3(request, measure, {}, callback);
 }
 static Return<ErrorStatus> ExecutePreparedModel(const sp<IPreparedModel>& preparedModel,
                                                 const Request& request, MeasureTiming measure,
@@ -461,7 +460,7 @@
                                                 Timing* timing) {
     ErrorStatus result;
     Return<void> ret = preparedModel->executeSynchronously_1_3(
-            request, measure,
+            request, measure, {},
             [&result, outputShapes, timing](ErrorStatus error, const hidl_vec<OutputShape>& shapes,
                                             const Timing& time) {
                 result = error;
@@ -716,7 +715,8 @@
         } break;
         case TestKind::QUANTIZATION_COUPLING: {
             ASSERT_TRUE(testModel.hasQuant8CoupledOperands());
-            createPreparedModel(device, model, &preparedModel, /*reportSkipping*/ false);
+            createPreparedModel(device, model, &preparedModel,
+                                /*reportSkipping*/ false);
             TestModel signedQuantizedModel = convertQuant8AsymmOperandsToSigned(testModel);
             sp<IPreparedModel> preparedCoupledModel;
             createPreparedModel(device, createModel(signedQuantizedModel), &preparedCoupledModel,
@@ -745,6 +745,12 @@
 void GeneratedTestBase::SetUp() {
     testing::TestWithParam<GeneratedTestParam>::SetUp();
     ASSERT_NE(kDevice, nullptr);
+
+    const Return<void> ret =
+            kDevice->supportsDeadlines([this](bool prepareModelDeadline, bool executionDeadline) {
+                mSupportsDeadlines = {prepareModelDeadline, executionDeadline};
+            });
+    ASSERT_TRUE(ret.isOk());
 }
 
 std::vector<NamedModel> getNamedModels(const FilterFn& filter) {
diff --git a/neuralnetworks/1.3/vts/functional/GeneratedTestHarness.h b/neuralnetworks/1.3/vts/functional/GeneratedTestHarness.h
index 2273e3b..fe695b4 100644
--- a/neuralnetworks/1.3/vts/functional/GeneratedTestHarness.h
+++ b/neuralnetworks/1.3/vts/functional/GeneratedTestHarness.h
@@ -36,6 +36,7 @@
     void SetUp() override;
     const sp<IDevice> kDevice = getData(std::get<NamedDevice>(GetParam()));
     const test_helper::TestModel& kTestModel = *getData(std::get<NamedModel>(GetParam()));
+    std::pair<bool, bool> mSupportsDeadlines;
 };
 
 using FilterFn = std::function<bool(const test_helper::TestModel&)>;
diff --git a/neuralnetworks/1.3/vts/functional/QualityOfServiceTests.cpp b/neuralnetworks/1.3/vts/functional/QualityOfServiceTests.cpp
new file mode 100644
index 0000000..62ffcda
--- /dev/null
+++ b/neuralnetworks/1.3/vts/functional/QualityOfServiceTests.cpp
@@ -0,0 +1,299 @@
+/*
+ * Copyright (C) 2019 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "1.0/Utils.h"
+#include "1.3/Callbacks.h"
+#include "1.3/Utils.h"
+#include "GeneratedTestHarness.h"
+#include "Utils.h"
+
+namespace android::hardware::neuralnetworks::V1_3::vts::functional {
+
+using implementation::ExecutionCallback;
+using implementation::PreparedModelCallback;
+using test_helper::TestBuffer;
+using test_helper::TestModel;
+using V1_1::ExecutionPreference;
+using V1_2::MeasureTiming;
+using V1_2::OutputShape;
+using V1_2::Timing;
+
+using HidlToken =
+        hidl_array<uint8_t, static_cast<uint32_t>(V1_2::Constant::BYTE_SIZE_OF_CACHE_TOKEN)>;
+
+enum class DeadlineBoundType { NOW, UNLIMITED };
+constexpr std::array<DeadlineBoundType, 2> deadlineBounds = {DeadlineBoundType::NOW,
+                                                             DeadlineBoundType::UNLIMITED};
+std::string toString(DeadlineBoundType type) {
+    switch (type) {
+        case DeadlineBoundType::NOW:
+            return "NOW";
+        case DeadlineBoundType::UNLIMITED:
+            return "UNLIMITED";
+    }
+    LOG(FATAL) << "Unrecognized DeadlineBoundType: " << static_cast<int>(type);
+    return {};
+}
+
+using Results = std::tuple<ErrorStatus, hidl_vec<OutputShape>, Timing>;
+using MaybeResults = std::optional<Results>;
+
+using ExecutionFunction =
+        std::function<MaybeResults(const sp<IPreparedModel>& preparedModel, const Request& request,
+                                   DeadlineBoundType deadlineBound)>;
+
+static OptionalTimePoint makeOptionalTimePoint(DeadlineBoundType deadlineBoundType) {
+    OptionalTimePoint deadline;
+    switch (deadlineBoundType) {
+        case DeadlineBoundType::NOW: {
+            const auto currentTime = std::chrono::steady_clock::now();
+            const auto currentTimeInNanoseconds =
+                    std::chrono::time_point_cast<std::chrono::nanoseconds>(currentTime);
+            const uint64_t nanosecondsSinceEpoch =
+                    currentTimeInNanoseconds.time_since_epoch().count();
+            deadline.nanoseconds(nanosecondsSinceEpoch);
+        } break;
+        case DeadlineBoundType::UNLIMITED: {
+            uint64_t unlimited = std::numeric_limits<uint64_t>::max();
+            deadline.nanoseconds(unlimited);
+        } break;
+    }
+    return deadline;
+}
+
+void runPrepareModelTest(const sp<IDevice>& device, const Model& model, Priority priority,
+                         std::optional<DeadlineBoundType> deadlineBound) {
+    OptionalTimePoint deadline;
+    if (deadlineBound.has_value()) {
+        deadline = makeOptionalTimePoint(deadlineBound.value());
+    }
+
+    // see if service can handle model
+    bool fullySupportsModel = false;
+    const Return<void> supportedCall = device->getSupportedOperations_1_3(
+            model, [&fullySupportsModel](ErrorStatus status, const hidl_vec<bool>& supported) {
+                ASSERT_EQ(ErrorStatus::NONE, status);
+                ASSERT_NE(0ul, supported.size());
+                fullySupportsModel = std::all_of(supported.begin(), supported.end(),
+                                                 [](bool valid) { return valid; });
+            });
+    ASSERT_TRUE(supportedCall.isOk());
+
+    // launch prepare model
+    const sp<PreparedModelCallback> preparedModelCallback = new PreparedModelCallback();
+    const Return<ErrorStatus> prepareLaunchStatus = device->prepareModel_1_3(
+            model, ExecutionPreference::FAST_SINGLE_ANSWER, priority, deadline,
+            hidl_vec<hidl_handle>(), hidl_vec<hidl_handle>(), HidlToken(), preparedModelCallback);
+    ASSERT_TRUE(prepareLaunchStatus.isOk());
+    ASSERT_EQ(ErrorStatus::NONE, static_cast<ErrorStatus>(prepareLaunchStatus));
+
+    // retrieve prepared model
+    preparedModelCallback->wait();
+    const ErrorStatus prepareReturnStatus = preparedModelCallback->getStatus();
+    const sp<V1_0::IPreparedModel> preparedModelV1_0 = preparedModelCallback->getPreparedModel();
+    const sp<IPreparedModel> preparedModel =
+            IPreparedModel::castFrom(preparedModelV1_0).withDefault(nullptr);
+
+    // The getSupportedOperations_1_3 call returns a list of operations that are
+    // guaranteed not to fail if prepareModel_1_3 is called, and
+    // 'fullySupportsModel' is true i.f.f. the entire model is guaranteed.
+    // If a driver has any doubt that it can prepare an operation, it must
+    // return false. So here, if a driver isn't sure if it can support an
+    // operation, but reports that it successfully prepared the model, the test
+    // can continue.
+    if (!fullySupportsModel && prepareReturnStatus != ErrorStatus::NONE) {
+        ASSERT_EQ(nullptr, preparedModel.get());
+        return;
+    }
+
+    // verify return status
+    if (!deadlineBound.has_value()) {
+        EXPECT_EQ(ErrorStatus::NONE, prepareReturnStatus);
+    } else {
+        switch (deadlineBound.value()) {
+            case DeadlineBoundType::NOW:
+                // If the execution was launched with a deadline of NOW, the
+                // deadline has already passed when the driver would launch the
+                // execution. In this case, the driver must return
+                // MISSED_DEADLINE_*.
+                EXPECT_TRUE(prepareReturnStatus == ErrorStatus::MISSED_DEADLINE_TRANSIENT ||
+                            prepareReturnStatus == ErrorStatus::MISSED_DEADLINE_PERSISTENT);
+                break;
+            case DeadlineBoundType::UNLIMITED:
+                // If an unlimited deadline is supplied, we expect the execution to
+                // proceed normally. In this case, check it normally by breaking out
+                // of the switch statement.
+                EXPECT_EQ(ErrorStatus::NONE, prepareReturnStatus);
+                break;
+        }
+    }
+    ASSERT_EQ(prepareReturnStatus == ErrorStatus::NONE, preparedModel.get() != nullptr);
+}
+
+void runPrepareModelTests(const sp<IDevice>& device, const Model& model,
+                          bool supportsPrepareModelDeadline) {
+    // test priority
+    for (auto priority : hidl_enum_range<Priority>{}) {
+        SCOPED_TRACE("priority: " + toString(priority));
+        if (priority == kDefaultPriority) continue;
+        runPrepareModelTest(device, model, priority, {});
+    }
+
+    // test deadline
+    if (supportsPrepareModelDeadline) {
+        for (auto deadlineBound : deadlineBounds) {
+            SCOPED_TRACE("deadlineBound: " + toString(deadlineBound));
+            runPrepareModelTest(device, model, kDefaultPriority, deadlineBound);
+        }
+    }
+}
+
+static MaybeResults executeAsynchronously(const sp<IPreparedModel>& preparedModel,
+                                          const Request& request, DeadlineBoundType deadlineBound) {
+    SCOPED_TRACE("asynchronous");
+    const MeasureTiming measure = MeasureTiming::NO;
+    const OptionalTimePoint deadline = makeOptionalTimePoint(deadlineBound);
+
+    // launch execution
+    const sp<ExecutionCallback> callback = new ExecutionCallback();
+    Return<ErrorStatus> ret = preparedModel->execute_1_3(request, measure, deadline, callback);
+    EXPECT_TRUE(ret.isOk());
+    EXPECT_EQ(ErrorStatus::NONE, ret.withDefault(ErrorStatus::GENERAL_FAILURE));
+    if (!ret.isOk() || ret != ErrorStatus::NONE) return std::nullopt;
+
+    // retrieve execution results
+    callback->wait();
+    const ErrorStatus status = callback->getStatus();
+    hidl_vec<OutputShape> outputShapes = callback->getOutputShapes();
+    const Timing timing = callback->getTiming();
+
+    // return results
+    return Results{status, std::move(outputShapes), timing};
+}
+
+static MaybeResults executeSynchronously(const sp<IPreparedModel>& preparedModel,
+                                         const Request& request, DeadlineBoundType deadlineBound) {
+    SCOPED_TRACE("synchronous");
+    const MeasureTiming measure = MeasureTiming::NO;
+    const OptionalTimePoint deadline = makeOptionalTimePoint(deadlineBound);
+
+    // configure results callback
+    MaybeResults results;
+    const auto cb = [&results](const auto&... args) { *results = {args...}; };
+
+    // run execution
+    const Return<void> ret =
+            preparedModel->executeSynchronously_1_3(request, measure, deadline, cb);
+    EXPECT_TRUE(ret.isOk());
+    if (!ret.isOk()) return std::nullopt;
+
+    // return results
+    return results;
+}
+
+void runExecutionTest(const sp<IPreparedModel>& preparedModel, const TestModel& testModel,
+                      const Request& request, bool synchronous, DeadlineBoundType deadlineBound) {
+    const ExecutionFunction execute = synchronous ? executeSynchronously : executeAsynchronously;
+
+    // Perform execution and unpack results.
+    const auto results = execute(preparedModel, request, deadlineBound);
+    if (!results.has_value()) return;
+    const auto& [status, outputShapes, timing] = results.value();
+
+    // Verify no timing information was returned
+    EXPECT_EQ(UINT64_MAX, timing.timeOnDevice);
+    EXPECT_EQ(UINT64_MAX, timing.timeInDriver);
+
+    // Validate deadline information if applicable.
+    switch (deadlineBound) {
+        case DeadlineBoundType::NOW:
+            // If the execution was launched with a deadline of NOW, the
+            // deadline has already passed when the driver would launch the
+            // execution. In this case, the driver must return
+            // MISSED_DEADLINE_*.
+            ASSERT_TRUE(status == ErrorStatus::MISSED_DEADLINE_TRANSIENT ||
+                        status == ErrorStatus::MISSED_DEADLINE_PERSISTENT);
+            return;
+        case DeadlineBoundType::UNLIMITED:
+            // If an unlimited deadline is supplied, we expect the execution to
+            // proceed normally. In this case, check it normally by breaking out
+            // of the switch statement.
+            ASSERT_EQ(ErrorStatus::NONE, status);
+            break;
+    }
+
+    // If the model output operands are fully specified, outputShapes must be either
+    // either empty, or have the same number of elements as the number of outputs.
+    ASSERT_TRUE(outputShapes.size() == 0 || outputShapes.size() == testModel.outputIndexes.size());
+
+    // Go through all outputs, check returned output shapes.
+    for (uint32_t i = 0; i < outputShapes.size(); i++) {
+        EXPECT_TRUE(outputShapes[i].isSufficient);
+        const auto& expect = testModel.operands[testModel.outputIndexes[i]].dimensions;
+        const std::vector<uint32_t> actual = outputShapes[i].dimensions;
+        EXPECT_EQ(expect, actual);
+    }
+
+    // Retrieve execution results.
+    ASSERT_TRUE(nn::compliantWithV1_0(request));
+    const V1_0::Request request10 = nn::convertToV1_0(request);
+    const std::vector<TestBuffer> outputs = getOutputBuffers(request10);
+
+    // We want "close-enough" results.
+    checkResults(testModel, outputs);
+}
+
+void runExecutionTests(const sp<IPreparedModel>& preparedModel, const TestModel& testModel,
+                       const Request& request) {
+    for (bool synchronous : {false, true}) {
+        for (auto deadlineBound : deadlineBounds) {
+            runExecutionTest(preparedModel, testModel, request, synchronous, deadlineBound);
+        }
+    }
+}
+
+void runTests(const sp<IDevice>& device, const TestModel& testModel,
+              std::pair<bool, bool> supportsDeadlines) {
+    // setup
+    const auto [supportsPrepareModelDeadline, supportsExecutionDeadline] = supportsDeadlines;
+    if (!supportsPrepareModelDeadline && !supportsExecutionDeadline) return;
+    const Model model = createModel(testModel);
+
+    // run prepare model tests
+    runPrepareModelTests(device, model, supportsPrepareModelDeadline);
+
+    if (supportsExecutionDeadline) {
+        // prepare model
+        sp<IPreparedModel> preparedModel;
+        createPreparedModel(device, model, &preparedModel);
+        if (preparedModel == nullptr) return;
+
+        // run execution tests
+        const Request request = nn::convertToV1_3(createRequest(testModel));
+        runExecutionTests(preparedModel, testModel, request);
+    }
+}
+
+class DeadlineTest : public GeneratedTestBase {};
+
+TEST_P(DeadlineTest, Test) {
+    runTests(kDevice, kTestModel, mSupportsDeadlines);
+}
+
+INSTANTIATE_GENERATED_TEST(DeadlineTest,
+                           [](const TestModel& testModel) { return !testModel.expectFailure; });
+
+}  // namespace android::hardware::neuralnetworks::V1_3::vts::functional
diff --git a/neuralnetworks/1.3/vts/functional/Utils.cpp b/neuralnetworks/1.3/vts/functional/Utils.cpp
new file mode 100644
index 0000000..23e2af8
--- /dev/null
+++ b/neuralnetworks/1.3/vts/functional/Utils.cpp
@@ -0,0 +1,27 @@
+/*
+ * Copyright (C) 2019 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "1.3/Utils.h"
+
+#include <iostream>
+
+namespace android::hardware::neuralnetworks::V1_3 {
+
+::std::ostream& operator<<(::std::ostream& os, ErrorStatus errorStatus) {
+    return os << toString(errorStatus);
+}
+
+}  // namespace android::hardware::neuralnetworks::V1_3
diff --git a/neuralnetworks/1.3/vts/functional/ValidateBurst.cpp b/neuralnetworks/1.3/vts/functional/ValidateBurst.cpp
index 7df8046..6ff9dfd 100644
--- a/neuralnetworks/1.3/vts/functional/ValidateBurst.cpp
+++ b/neuralnetworks/1.3/vts/functional/ValidateBurst.cpp
@@ -34,7 +34,6 @@
 using nn::ExecutionBurstController;
 using nn::RequestChannelSender;
 using nn::ResultChannelReceiver;
-using V1_0::ErrorStatus;
 using V1_0::Request;
 using V1_2::FmqRequestDatum;
 using V1_2::FmqResultDatum;
@@ -80,16 +79,17 @@
     ASSERT_NE(nullptr, fmqResultDescriptor);
 
     // configure burst
-    ErrorStatus errorStatus;
+    V1_0::ErrorStatus errorStatus;
     sp<IBurstContext> burstContext;
     const Return<void> ret = preparedModel->configureExecutionBurst(
             callback, *fmqRequestDescriptor, *fmqResultDescriptor,
-            [&errorStatus, &burstContext](ErrorStatus status, const sp<IBurstContext>& context) {
+            [&errorStatus, &burstContext](V1_0::ErrorStatus status,
+                                          const sp<IBurstContext>& context) {
                 errorStatus = status;
                 burstContext = context;
             });
     ASSERT_TRUE(ret.isOk());
-    ASSERT_EQ(ErrorStatus::NONE, errorStatus);
+    ASSERT_EQ(V1_0::ErrorStatus::NONE, errorStatus);
     ASSERT_NE(nullptr, burstContext.get());
 
     // return values
@@ -144,7 +144,7 @@
     auto results = receiver->getBlocking();
     ASSERT_TRUE(results.has_value());
     const auto [status, outputShapes, timing] = std::move(*results);
-    EXPECT_NE(ErrorStatus::NONE, status);
+    EXPECT_NE(V1_0::ErrorStatus::NONE, status);
     EXPECT_EQ(0u, outputShapes.size());
     EXPECT_TRUE(badTiming(timing));
 }
@@ -302,14 +302,15 @@
     // collect serialized result by running regular burst
     const auto [nRegular, outputShapesRegular, timingRegular, fallbackRegular] =
             controllerRegular->compute(request, MeasureTiming::NO, keys);
-    const ErrorStatus statusRegular = nn::convertResultCodeToErrorStatus(nRegular);
+    const V1_0::ErrorStatus statusRegular =
+            nn::convertToV1_0(nn::convertResultCodeToErrorStatus(nRegular));
     EXPECT_FALSE(fallbackRegular);
 
     // skip test if regular burst output isn't useful for testing a failure
     // caused by having too small of a length for the result FMQ
     const std::vector<FmqResultDatum> serialized =
             android::nn::serialize(statusRegular, outputShapesRegular, timingRegular);
-    if (statusRegular != ErrorStatus::NONE ||
+    if (statusRegular != V1_0::ErrorStatus::NONE ||
         serialized.size() <= kExecutionBurstChannelSmallLength) {
         return;
     }
@@ -318,8 +319,9 @@
     // large enough to return the serialized result
     const auto [nSmall, outputShapesSmall, timingSmall, fallbackSmall] =
             controllerSmall->compute(request, MeasureTiming::NO, keys);
-    const ErrorStatus statusSmall = nn::convertResultCodeToErrorStatus(nSmall);
-    EXPECT_NE(ErrorStatus::NONE, statusSmall);
+    const V1_0::ErrorStatus statusSmall =
+            nn::convertToV1_0(nn::convertResultCodeToErrorStatus(nSmall));
+    EXPECT_NE(V1_0::ErrorStatus::NONE, statusSmall);
     EXPECT_EQ(0u, outputShapesSmall.size());
     EXPECT_TRUE(badTiming(timingSmall));
     EXPECT_FALSE(fallbackSmall);
diff --git a/neuralnetworks/1.3/vts/functional/ValidateModel.cpp b/neuralnetworks/1.3/vts/functional/ValidateModel.cpp
index cc86264..a211428 100644
--- a/neuralnetworks/1.3/vts/functional/ValidateModel.cpp
+++ b/neuralnetworks/1.3/vts/functional/ValidateModel.cpp
@@ -18,13 +18,13 @@
 
 #include "1.0/Utils.h"
 #include "1.3/Callbacks.h"
+#include "1.3/Utils.h"
 #include "GeneratedTestHarness.h"
 #include "VtsHalNeuralnetworks.h"
 
 namespace android::hardware::neuralnetworks::V1_3::vts::functional {
 
 using implementation::PreparedModelCallback;
-using V1_0::ErrorStatus;
 using V1_1::ExecutionPreference;
 using V1_2::SymmPerChannelQuantParams;
 using HidlToken =
@@ -44,13 +44,19 @@
 }
 
 static void validatePrepareModel(const sp<IDevice>& device, const std::string& message,
-                                 const Model& model, ExecutionPreference preference) {
+                                 const Model& model, ExecutionPreference preference,
+                                 bool testDeadline) {
     SCOPED_TRACE(message + " [prepareModel_1_3]");
 
+    OptionalTimePoint deadline;
+    if (testDeadline) {
+        deadline.nanoseconds(std::numeric_limits<uint64_t>::max());
+    }
+
     sp<PreparedModelCallback> preparedModelCallback = new PreparedModelCallback();
-    Return<ErrorStatus> prepareLaunchStatus =
-            device->prepareModel_1_3(model, preference, hidl_vec<hidl_handle>(),
-                                     hidl_vec<hidl_handle>(), HidlToken(), preparedModelCallback);
+    Return<ErrorStatus> prepareLaunchStatus = device->prepareModel_1_3(
+            model, preference, kDefaultPriority, deadline, hidl_vec<hidl_handle>(),
+            hidl_vec<hidl_handle>(), HidlToken(), preparedModelCallback);
     ASSERT_TRUE(prepareLaunchStatus.isOk());
     ASSERT_EQ(ErrorStatus::INVALID_ARGUMENT, static_cast<ErrorStatus>(prepareLaunchStatus));
 
@@ -73,12 +79,13 @@
 // to the model does not leave this function.
 static void validate(const sp<IDevice>& device, const std::string& message, Model model,
                      const std::function<void(Model*)>& mutation,
-                     ExecutionPreference preference = ExecutionPreference::FAST_SINGLE_ANSWER) {
+                     ExecutionPreference preference = ExecutionPreference::FAST_SINGLE_ANSWER,
+                     bool testDeadline = false) {
     mutation(&model);
-    if (validExecutionPreference(preference)) {
+    if (validExecutionPreference(preference) && !testDeadline) {
         validateGetSupportedOperations(device, message, model);
     }
-    validatePrepareModel(device, message, model, preference);
+    validatePrepareModel(device, message, model, preference, testDeadline);
 }
 
 static uint32_t addOperand(Model* model) {
@@ -714,9 +721,19 @@
     }
 }
 
+///////////////////////// DEADLINE /////////////////////////
+
+static void deadlineTest(const sp<IDevice>& device, const Model& model) {
+    const std::string message = "deadlineTest: deadline not supported";
+    const auto noop = [](Model*) {};
+    validate(device, message, model, noop, ExecutionPreference::FAST_SINGLE_ANSWER,
+             /*testDeadline=*/true);
+}
+
 ////////////////////////// ENTRY POINT //////////////////////////////
 
-void validateModel(const sp<IDevice>& device, const Model& model) {
+void validateModel(const sp<IDevice>& device, const Model& model,
+                   bool prepareModelDeadlineSupported) {
     mutateOperandTypeTest(device, model);
     mutateOperandRankTest(device, model);
     mutateOperandScaleTest(device, model);
@@ -732,6 +749,9 @@
     addOperationInputTest(device, model);
     addOperationOutputTest(device, model);
     mutateExecutionPreferenceTest(device, model);
+    if (!prepareModelDeadlineSupported) {
+        deadlineTest(device, model);
+    }
 }
 
 }  // namespace android::hardware::neuralnetworks::V1_3::vts::functional
diff --git a/neuralnetworks/1.3/vts/functional/ValidateRequest.cpp b/neuralnetworks/1.3/vts/functional/ValidateRequest.cpp
index 96dc589..be4112a 100644
--- a/neuralnetworks/1.3/vts/functional/ValidateRequest.cpp
+++ b/neuralnetworks/1.3/vts/functional/ValidateRequest.cpp
@@ -18,7 +18,7 @@
 
 #include <chrono>
 #include "1.0/Utils.h"
-#include "1.2/Callbacks.h"
+#include "1.3/Callbacks.h"
 #include "ExecutionBurstController.h"
 #include "GeneratedTestHarness.h"
 #include "TestHarness.h"
@@ -27,11 +27,10 @@
 
 namespace android::hardware::neuralnetworks::V1_3::vts::functional {
 
-using V1_0::ErrorStatus;
+using implementation::ExecutionCallback;
 using V1_2::MeasureTiming;
 using V1_2::OutputShape;
 using V1_2::Timing;
-using V1_2::implementation::ExecutionCallback;
 
 ///////////////////////// UTILITY FUNCTIONS /////////////////////////
 
@@ -44,7 +43,8 @@
 // that use the request. Note that the request here is passed by value, and any
 // mutation to the request does not leave this function.
 static void validate(const sp<IPreparedModel>& preparedModel, const std::string& message,
-                     Request request, const std::function<void(Request*)>& mutation) {
+                     Request request, const std::function<void(Request*)>& mutation,
+                     bool testDeadline = false) {
     mutation(&request);
 
     // We'd like to test both with timing requested and without timing
@@ -57,13 +57,18 @@
     };
     MeasureTiming measure = (hash & 1) ? MeasureTiming::YES : MeasureTiming::NO;
 
+    OptionalTimePoint deadline;
+    if (testDeadline) {
+        deadline.nanoseconds(std::numeric_limits<uint64_t>::max());
+    }
+
     // asynchronous
     {
         SCOPED_TRACE(message + " [execute_1_3]");
 
         sp<ExecutionCallback> executionCallback = new ExecutionCallback();
         Return<ErrorStatus> executeLaunchStatus =
-                preparedModel->execute_1_3(request, measure, executionCallback);
+                preparedModel->execute_1_3(request, measure, deadline, executionCallback);
         ASSERT_TRUE(executeLaunchStatus.isOk());
         ASSERT_EQ(ErrorStatus::INVALID_ARGUMENT, static_cast<ErrorStatus>(executeLaunchStatus));
 
@@ -81,7 +86,7 @@
         SCOPED_TRACE(message + " [executeSynchronously_1_3]");
 
         Return<void> executeStatus = preparedModel->executeSynchronously_1_3(
-                request, measure,
+                request, measure, deadline,
                 [](ErrorStatus error, const hidl_vec<OutputShape>& outputShapes,
                    const Timing& timing) {
                     ASSERT_EQ(ErrorStatus::INVALID_ARGUMENT, error);
@@ -93,7 +98,7 @@
 
     // burst
     // TODO(butlermichael): Check if we need to test burst in V1_3 if the interface remains V1_2.
-    {
+    if (!testDeadline) {
         SCOPED_TRACE(message + " [burst]");
 
         ASSERT_TRUE(nn::compliantWithV1_0(request));
@@ -153,17 +158,29 @@
     }
 }
 
+///////////////////////// DEADLINE ////////////////////////////////////
+
+static void deadlineTest(const sp<IPreparedModel>& preparedModel, const Request& request) {
+    const std::string message = "deadlineTest: deadline not supported";
+    const auto noop = [](Request*) {};
+    validate(preparedModel, message, request, noop, /*testDeadline=*/true);
+}
+
 ///////////////////////////// ENTRY POINT //////////////////////////////////
 
-void validateRequest(const sp<IPreparedModel>& preparedModel, const Request& request) {
+void validateRequest(const sp<IPreparedModel>& preparedModel, const Request& request,
+                     bool executionDeadlineSupported) {
     removeInputTest(preparedModel, request);
     removeOutputTest(preparedModel, request);
+    if (!executionDeadlineSupported) {
+        deadlineTest(preparedModel, request);
+    }
 }
 
 void validateRequestFailure(const sp<IPreparedModel>& preparedModel, const Request& request) {
     SCOPED_TRACE("Expecting request to fail [executeSynchronously_1_3]");
     Return<void> executeStatus = preparedModel->executeSynchronously_1_3(
-            request, MeasureTiming::NO,
+            request, MeasureTiming::NO, {},
             [](ErrorStatus error, const hidl_vec<OutputShape>& outputShapes, const Timing& timing) {
                 ASSERT_NE(ErrorStatus::NONE, error);
                 EXPECT_EQ(outputShapes.size(), 0);
diff --git a/neuralnetworks/1.3/vts/functional/VtsHalNeuralnetworks.cpp b/neuralnetworks/1.3/vts/functional/VtsHalNeuralnetworks.cpp
index 1140b68..93c8f13 100644
--- a/neuralnetworks/1.3/vts/functional/VtsHalNeuralnetworks.cpp
+++ b/neuralnetworks/1.3/vts/functional/VtsHalNeuralnetworks.cpp
@@ -23,6 +23,7 @@
 #include <utility>
 #include "1.0/Utils.h"
 #include "1.3/Callbacks.h"
+#include "1.3/Utils.h"
 #include "GeneratedTestHarness.h"
 #include "TestHarness.h"
 #include "Utils.h"
@@ -32,7 +33,6 @@
 using HidlToken =
         hidl_array<uint8_t, static_cast<uint32_t>(V1_2::Constant::BYTE_SIZE_OF_CACHE_TOKEN)>;
 using implementation::PreparedModelCallback;
-using V1_0::ErrorStatus;
 using V1_1::ExecutionPreference;
 
 // internal helper function
@@ -55,8 +55,8 @@
     // launch prepare model
     const sp<PreparedModelCallback> preparedModelCallback = new PreparedModelCallback();
     const Return<ErrorStatus> prepareLaunchStatus = device->prepareModel_1_3(
-            model, ExecutionPreference::FAST_SINGLE_ANSWER, hidl_vec<hidl_handle>(),
-            hidl_vec<hidl_handle>(), HidlToken(), preparedModelCallback);
+            model, ExecutionPreference::FAST_SINGLE_ANSWER, kDefaultPriority, {},
+            hidl_vec<hidl_handle>(), hidl_vec<hidl_handle>(), HidlToken(), preparedModelCallback);
     ASSERT_TRUE(prepareLaunchStatus.isOk());
     ASSERT_EQ(ErrorStatus::NONE, static_cast<ErrorStatus>(prepareLaunchStatus));
 
@@ -84,6 +84,7 @@
                   << std::endl;
         GTEST_SKIP();
     }
+
     ASSERT_EQ(ErrorStatus::NONE, prepareReturnStatus);
     ASSERT_NE(nullptr, preparedModel->get());
 }
@@ -122,23 +123,27 @@
 INSTANTIATE_DEVICE_TEST(NeuralnetworksHidlTest);
 
 // Forward declaration from ValidateModel.cpp
-void validateModel(const sp<IDevice>& device, const Model& model);
+void validateModel(const sp<IDevice>& device, const Model& model,
+                   bool prepareModelDeadlineSupported);
 // Forward declaration from ValidateRequest.cpp
-void validateRequest(const sp<IPreparedModel>& preparedModel, const Request& request);
+void validateRequest(const sp<IPreparedModel>& preparedModel, const Request& request,
+                     bool executionDeadlineSupported);
 // Forward declaration from ValidateRequest.cpp
 void validateRequestFailure(const sp<IPreparedModel>& preparedModel, const Request& request);
 // Forward declaration from ValidateBurst.cpp
 void validateBurst(const sp<IPreparedModel>& preparedModel, const V1_0::Request& request);
 
-void validateEverything(const sp<IDevice>& device, const Model& model, const Request& request) {
-    validateModel(device, model);
+void validateEverything(const sp<IDevice>& device, const Model& model, const Request& request,
+                        std::pair<bool, bool> supportsDeadlines) {
+    const auto [prepareModelDeadlineSupported, executionDeadlineSupported] = supportsDeadlines;
+    validateModel(device, model, prepareModelDeadlineSupported);
 
     // Create IPreparedModel.
     sp<IPreparedModel> preparedModel;
     createPreparedModel(device, model, &preparedModel);
     if (preparedModel == nullptr) return;
 
-    validateRequest(preparedModel, request);
+    validateRequest(preparedModel, request, executionDeadlineSupported);
 
     // TODO(butlermichael): Check if we need to test burst in V1_3 if the interface remains V1_2.
     ASSERT_TRUE(nn::compliantWithV1_0(request));
@@ -146,10 +151,12 @@
     validateBurst(preparedModel, request10);
 }
 
-void validateFailure(const sp<IDevice>& device, const Model& model, const Request& request) {
+void validateFailure(const sp<IDevice>& device, const Model& model, const Request& request,
+                     std::pair<bool, bool> supportsDeadlines) {
+    const bool prepareModelDeadlineSupported = supportsDeadlines.first;
     // TODO: Should this always succeed?
     //       What if the invalid input is part of the model (i.e., a parameter).
-    validateModel(device, model);
+    validateModel(device, model, prepareModelDeadlineSupported);
 
     // Create IPreparedModel.
     sp<IPreparedModel> preparedModel;
@@ -163,9 +170,9 @@
     const Model model = createModel(kTestModel);
     const Request request = nn::convertToV1_3(createRequest(kTestModel));
     if (kTestModel.expectFailure) {
-        validateFailure(kDevice, model, request);
+        validateFailure(kDevice, model, request, mSupportsDeadlines);
     } else {
-        validateEverything(kDevice, model, request);
+        validateEverything(kDevice, model, request, mSupportsDeadlines);
     }
 }
 
diff --git a/neuralnetworks/1.3/vts/functional/include/1.3/Callbacks.h b/neuralnetworks/1.3/vts/functional/include/1.3/Callbacks.h
index fb19a84..e9dec2d 100644
--- a/neuralnetworks/1.3/vts/functional/include/1.3/Callbacks.h
+++ b/neuralnetworks/1.3/vts/functional/include/1.3/Callbacks.h
@@ -18,8 +18,11 @@
 #define ANDROID_HARDWARE_NEURALNETWORKS_V1_3_CALLBACKS_H
 
 #include <android-base/thread_annotations.h>
+#include <android/hardware/neuralnetworks/1.0/IExecutionCallback.h>
 #include <android/hardware/neuralnetworks/1.0/IPreparedModelCallback.h>
+#include <android/hardware/neuralnetworks/1.2/IExecutionCallback.h>
 #include <android/hardware/neuralnetworks/1.2/IPreparedModelCallback.h>
+#include <android/hardware/neuralnetworks/1.3/IExecutionCallback.h>
 #include <android/hardware/neuralnetworks/1.3/IPreparedModelCallback.h>
 #include <hidl/Status.h>
 #include <condition_variable>
@@ -136,7 +139,7 @@
      * @param preparedModel Returned model that has been prepared for execution,
      *     nullptr if the model was unable to be prepared.
      */
-    Return<void> notify_1_3(V1_0::ErrorStatus status,
+    Return<void> notify_1_3(V1_3::ErrorStatus status,
                             const sp<V1_3::IPreparedModel>& preparedModel) override;
 
     /**
@@ -158,7 +161,7 @@
      *     - GENERAL_FAILURE if there is an unspecified error
      *     - INVALID_ARGUMENT if the input model is invalid
      */
-    V1_0::ErrorStatus getStatus() const;
+    ErrorStatus getStatus() const;
 
     /**
      * Retrieves the model that has been prepared for execution from the
@@ -173,13 +176,216 @@
     sp<V1_0::IPreparedModel> getPreparedModel() const;
 
   private:
+    Return<void> notifyInternal(ErrorStatus status, const sp<V1_0::IPreparedModel>& preparedModel);
+
     mutable std::mutex mMutex;
     mutable std::condition_variable mCondition;
     bool mNotified GUARDED_BY(mMutex) = false;
-    V1_0::ErrorStatus mErrorStatus = V1_0::ErrorStatus::GENERAL_FAILURE;
+    ErrorStatus mErrorStatus = ErrorStatus::GENERAL_FAILURE;
     sp<V1_0::IPreparedModel> mPreparedModel;
 };
 
+/**
+ * The ExecutionCallback class is used to receive the results of the execution
+ * from a task executing asynchronously with respect to the runtime. If a
+ * calling thread calls wait or get* on a ExecutionCallback object and the
+ * corresponding asynchronous task has not finished the execution, the calling
+ * thread will block until the asynchronous task has either called one of the
+ * notify* methods.
+ *
+ * If the callback object is notified more than once, only the results of the
+ * first call to notify* are used, and the results from subsequent calls are
+ * discarded.
+ *
+ * This callback object is passed as an argument to IPreparedModel::execute*.
+ */
+class ExecutionCallback : public IExecutionCallback {
+  public:
+    /**
+     * IExecutionCallback::notify marks the callback object with the return
+     * status of the asynchronous execution that held this callback and enables
+     * all prior and future wait calls on the ExecutionCallback object to
+     * proceed.
+     *
+     * One of the IExecutionCallback::notify* methods must be called on a given
+     * ExecutionCallback object.
+     *
+     * If the callback object is notified more than once, only the results of
+     * the first call to notify* are used, and the results from subsequent calls
+     * are discarded.
+     *
+     * @param status Error status returned from launching the asynchronous task
+     *     (if the launch fails) or from the asynchronous task itself (if the
+     *     launch succeeds). Must be:
+     *     - NONE if the asynchronous execution was successful
+     *     - DEVICE_UNAVAILABLE if driver is offline or busy
+     *     - GENERAL_FAILURE if there is an unspecified error
+     *     - OUTPUT_INSUFFICIENT_SIZE if provided output buffer is not large
+     *         enough to store the resultant values
+     *     - INVALID_ARGUMENT if the input request is invalid
+     */
+    Return<void> notify(V1_0::ErrorStatus status) override;
+
+    /**
+     * IExecutionCallback::notify_1_2 marks the callback object with the results
+     * (error status, dynamic output shapes, and timing information) of the
+     * asynchronous execution that held this callback and enables all prior and
+     * future wait calls on the ExecutionCallback object to proceed.
+     *
+     * One of the IExecutionCallback::notify* methods must be called on a given
+     * ExecutionCallback object.
+     *
+     * If the callback object is notified more than once, only the results of
+     * the first call to notify* are used, and the results from subsequent calls
+     * are discarded.
+     *
+     * @param status Error status returned from launching the asynchronous task
+     *     (if the launch fails) or from the asynchronous task itself (if the
+     *     launch succeeds). Must be:
+     *     - NONE if the asynchronous execution was successful
+     *     - DEVICE_UNAVAILABLE if driver is offline or busy
+     *     - GENERAL_FAILURE if the asynchronous task resulted in an unspecified
+     *         error
+     *     - OUTPUT_INSUFFICIENT_SIZE if at least one output operand buffer is
+     *         not large enough to store the corresponding output
+     *     - INVALID_ARGUMENT if one of the input arguments to prepareModel is
+     *         invalid
+     * @param outputShapes A list of shape information of model output operands.
+     *     The index into "outputShapes" corresponds to the index of the output
+     *     operand in the Request outputs vector. outputShapes must be empty
+     *     unless the status is either NONE or OUTPUT_INSUFFICIENT_SIZE.
+     * @param Timing Duration of execution. Unless MeasureTiming::YES was passed
+     *     when launching the execution and status is NONE, all times must be
+     *     reported as UINT64_MAX. A driver may choose to report any time as
+     *     UINT64_MAX, indicating that particular measurement is not available.
+     */
+    Return<void> notify_1_2(V1_0::ErrorStatus status,
+                            const hidl_vec<V1_2::OutputShape>& outputShapes,
+                            const V1_2::Timing& timing) override;
+
+    /**
+     * IExecutionCallback::notify_1_3 marks the callback object with the results
+     * (error status, dynamic output shapes, and timing information) of the
+     * asynchronous execution that held this callback and enables all prior and
+     * future wait calls on the ExecutionCallback object to proceed.
+     *
+     * One of the IExecutionCallback::notify* methods must be called on a given
+     * ExecutionCallback object.
+     *
+     * If the callback object is notified more than once, only the results of
+     * the first call to notify* are used, and the results from subsequent calls
+     * are discarded.
+     *
+     * @param status Error status returned from launching the asynchronous task
+     *     (if the launch fails) or from the asynchronous task itself (if the
+     *     launch succeeds). Must be:
+     *     - NONE if the asynchronous execution was successful
+     *     - DEVICE_UNAVAILABLE if driver is offline or busy
+     *     - GENERAL_FAILURE if the asynchronous task resulted in an unspecified
+     *         error
+     *     - OUTPUT_INSUFFICIENT_SIZE if at least one output operand buffer is
+     *         not large enough to store the corresponding output
+     *     - INVALID_ARGUMENT if one of the input arguments to prepareModel is
+     *         invalid
+     *     - MISSED_DEADLINE_* if the deadline was not met
+     * @param outputShapes A list of shape information of model output operands.
+     *     The index into "outputShapes" corresponds to the index of the output
+     *     operand in the Request outputs vector. outputShapes must be empty
+     *     unless the status is either NONE or OUTPUT_INSUFFICIENT_SIZE.
+     * @param Timing Duration of execution. Unless MeasureTiming::YES was passed
+     *     when launching the execution and status is NONE, all times must be
+     *     reported as UINT64_MAX. A driver may choose to report any time as
+     *     UINT64_MAX, indicating that particular measurement is not available.
+     */
+    Return<void> notify_1_3(V1_3::ErrorStatus status,
+                            const hidl_vec<V1_2::OutputShape>& outputShapes,
+                            const V1_2::Timing& timing) override;
+
+    /**
+     * ExecutionCallback::wait blocks until notify* has been called on the
+     * callback object.
+     */
+    void wait() const;
+
+    /**
+     * Retrieves the error status returned from the asynchronous task launched
+     * by one of the IPreparedModel::execute* methods. If
+     * IPreparedModel::execute* (but not IPreparedModel::executeSynchronously*)
+     * has not finished asynchronously executing, this call will block until the
+     * asynchronous task notifies the object.
+     *
+     * @return status Error status returned from launching the asynchronous task
+     *     (if the launch fails) or from the asynchronous task itself (if the
+     *     launch succeeds). Must be:
+     *     - NONE if the asynchronous execution was successful
+     *     - DEVICE_UNAVAILABLE if driver is offline or busy
+     *     - GENERAL_FAILURE if the asynchronous task resulted in an unspecified
+     *         error
+     *     - OUTPUT_INSUFFICIENT_SIZE if at least one output operand buffer is
+     *         not large enough to store the corresponding output
+     *     - INVALID_ARGUMENT if one of the input arguments to prepareModel is
+     *         invalid
+     *     - MISSED_DEADLINE_* if the deadline could not be met
+     */
+    V1_3::ErrorStatus getStatus() const;
+
+    /**
+     * Retrieves the error status returned from the asynchronous task launched
+     * by one of the IPreparedModel::execute* methods. If
+     * IPreparedModel::execute* (but not IPreparedModel::executeSynchronously*)
+     * has not finished asynchronously executing, this call will block until the
+     * asynchronous task notifies the object.
+     *
+     * If the asynchronous task was launched by IPreparedModel::execute, an
+     * empty vector will be returned.
+     *
+     * @return outputShapes A list of shape information of model output
+     *     operands. The index into "outputShapes" corresponds to the index of
+     *     the output operand in the Request outputs vector. outputShapes must
+     *     be empty unless the status is either NONE or
+     *     OUTPUT_INSUFFICIENT_SIZE. outputShaps may be empty if the status is
+     *     NONE and all model output operands are fully-specified at execution
+     *     time. outputShapes must have the same number of elements as the
+     *     number of model output operands if the status is
+     *     OUTPUT_INSUFFICIENT_SIZE, or if the status is NONE and the model has
+     *     at least one output operand that is not fully-specified.
+     */
+    const std::vector<V1_2::OutputShape>& getOutputShapes() const;
+
+    /**
+     * Retrieves the error status returned from the asynchronous task launched
+     * by one of the IPreparedModel::execute* methods. If
+     * IPreparedModel::execute* (but not IPreparedModel::executeSynchronously*)
+     * has not finished asynchronously executing, this call will block until the
+     * asynchronous task notifies the object.
+     *
+     * If the asynchronous task was launched by IPreparedModel::execute, every
+     * time must be UINT64_MAX.
+     *
+     * @return timing Duration of the execution. Every time must be UINT64_MAX
+     *     unless the status is NONE.
+     */
+    V1_2::Timing getTiming() const;
+
+  private:
+    /*
+     * ExecutionCallback::notifyInternal stores the results of the execution
+     * (status, output shapes, and timing information) in the ExecutionCallback
+     * object before any call to wait or get* return. It then enables all prior
+     * and future wait calls on the ExecutionCallback object to proceed.
+     */
+    Return<void> notifyInternal(V1_3::ErrorStatus errorStatus,
+                                hidl_vec<V1_2::OutputShape> outputShapes, V1_2::Timing timing);
+
+    // members
+    mutable std::mutex mMutex;
+    mutable std::condition_variable mCondition;
+    bool mNotified GUARDED_BY(mMutex) = false;
+    V1_3::ErrorStatus mErrorStatus = V1_3::ErrorStatus::GENERAL_FAILURE;
+    std::vector<V1_2::OutputShape> mOutputShapes = {};
+    V1_2::Timing mTiming = {};
+};
+
 }  // namespace android::hardware::neuralnetworks::V1_3::implementation
 
 #endif  // ANDROID_HARDWARE_NEURALNETWORKS_V1_3_CALLBACKS_H
diff --git a/neuralnetworks/1.3/vts/functional/include/1.3/Utils.h b/neuralnetworks/1.3/vts/functional/include/1.3/Utils.h
new file mode 100644
index 0000000..3661b66
--- /dev/null
+++ b/neuralnetworks/1.3/vts/functional/include/1.3/Utils.h
@@ -0,0 +1,36 @@
+/*
+ * Copyright (C) 2019 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef ANDROID_HARDWARE_NEURALNETWORKS_V1_3_UTILS_H
+#define ANDROID_HARDWARE_NEURALNETWORKS_V1_3_UTILS_H
+
+#include <android/hardware/neuralnetworks/1.3/types.h>
+#include <iosfwd>
+
+namespace android::hardware::neuralnetworks {
+
+inline constexpr V1_3::Priority kDefaultPriority = V1_3::Priority::MEDIUM;
+
+}  // namespace android::hardware::neuralnetworks
+
+namespace android::hardware::neuralnetworks::V1_3 {
+
+// pretty-print values for error messages
+::std::ostream& operator<<(::std::ostream& os, ErrorStatus errorStatus);
+
+}  // namespace android::hardware::neuralnetworks::V1_3
+
+#endif  // ANDROID_HARDWARE_NEURALNETWORKS_V1_3_UTILS_H