diff --git a/neuralnetworks/1.3/vts/functional/Android.bp b/neuralnetworks/1.3/vts/functional/Android.bp
index e7a9fd3..ce2d3a9 100644
--- a/neuralnetworks/1.3/vts/functional/Android.bp
+++ b/neuralnetworks/1.3/vts/functional/Android.bp
@@ -40,6 +40,7 @@
         "BasicTests.cpp",
         "CompilationCachingTests.cpp",
         "GeneratedTestHarness.cpp",
+        "QualityOfServiceTests.cpp",
         "TestAssertions.cpp",
         "ValidateBurst.cpp",
         "ValidateModel.cpp",
diff --git a/neuralnetworks/1.3/vts/functional/GeneratedTestHarness.cpp b/neuralnetworks/1.3/vts/functional/GeneratedTestHarness.cpp
index 82e63ac..a2c0c4e 100644
--- a/neuralnetworks/1.3/vts/functional/GeneratedTestHarness.cpp
+++ b/neuralnetworks/1.3/vts/functional/GeneratedTestHarness.cpp
@@ -45,6 +45,7 @@
 
 #include "1.0/Utils.h"
 #include "1.3/Callbacks.h"
+#include "1.3/Utils.h"
 #include "ExecutionBurstController.h"
 #include "MemoryUtils.h"
 #include "TestHarness.h"
@@ -714,7 +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,
@@ -743,6 +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/ValidateModel.cpp b/neuralnetworks/1.3/vts/functional/ValidateModel.cpp
index 43e53ef..a211428 100644
--- a/neuralnetworks/1.3/vts/functional/ValidateModel.cpp
+++ b/neuralnetworks/1.3/vts/functional/ValidateModel.cpp
@@ -44,12 +44,18 @@
 }
 
 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, kDefaultPriority, {}, hidl_vec<hidl_handle>(),
+            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 9fb4c6e..be4112a 100644
--- a/neuralnetworks/1.3/vts/functional/ValidateRequest.cpp
+++ b/neuralnetworks/1.3/vts/functional/ValidateRequest.cpp
@@ -43,7 +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
@@ -56,13 +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));
 
@@ -80,7 +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);
@@ -92,7 +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));
@@ -152,11 +158,23 @@
     }
 }
 
+///////////////////////// 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) {
diff --git a/neuralnetworks/1.3/vts/functional/VtsHalNeuralnetworks.cpp b/neuralnetworks/1.3/vts/functional/VtsHalNeuralnetworks.cpp
index 7a32b04..93c8f13 100644
--- a/neuralnetworks/1.3/vts/functional/VtsHalNeuralnetworks.cpp
+++ b/neuralnetworks/1.3/vts/functional/VtsHalNeuralnetworks.cpp
@@ -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);
     }
 }