diff --git a/neuralnetworks/1.3/Android.bp b/neuralnetworks/1.3/Android.bp
index 0b07a58..08e824d 100644
--- a/neuralnetworks/1.3/Android.bp
+++ b/neuralnetworks/1.3/Android.bp
@@ -8,6 +8,7 @@
     },
     srcs: [
         "types.hal",
+        "IBuffer.hal",
         "IDevice.hal",
         "IPreparedModel.hal",
         "IPreparedModelCallback.hal",
diff --git a/neuralnetworks/1.3/IBuffer.hal b/neuralnetworks/1.3/IBuffer.hal
new file mode 100644
index 0000000..84241c5
--- /dev/null
+++ b/neuralnetworks/1.3/IBuffer.hal
@@ -0,0 +1,57 @@
+/*
+ * Copyright (C) 2020 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 @1.0::ErrorStatus;
+
+/**
+ * This interface represents a device memory buffer.
+ */
+interface IBuffer {
+    /**
+     * Retrieves the content of this buffer to a shared memory region.
+     *
+     * The IBuffer object must have been initialized before the call to IBuffer::copyTo.
+     * For more information on the state of the IBuffer object, refer to IDevice::allocate.
+     *
+     * @param dst The destination shared memory region.
+     * @return status Error status of the call, must be:
+     *     - NONE if successful
+     *     - DEVICE_UNAVAILABLE if driver is offline or busy
+     *     - GENERAL_FAILURE if the IBuffer object is uninitialized, or there is an unspecified
+     *       error
+     *     - INVALID_ARGUMENT if provided memory is invalid
+     */
+    copyTo(memory dst) generates (ErrorStatus status);
+
+    /**
+     * Sets the content of this buffer from a shared memory region.
+     *
+     * @param src The source shared memory region.
+     * @param dimensions Updated dimensional information. If the dimensions of the IBuffer object
+     *     are not fully specified, then the dimensions must be fully specified here. If the
+     *     dimensions of the IBuffer object are fully specified, then the dimensions may be empty
+     *     here. If dimensions.size() > 0, then all dimensions must be specified here, and any
+     *     dimension that was specified in the IBuffer object must have the same value here.
+     * @return status Error status of the call, must be:
+     *     - NONE if successful
+     *     - DEVICE_UNAVAILABLE if driver is offline or busy
+     *     - GENERAL_FAILURE if there is an unspecified error
+     *     - INVALID_ARGUMENT if provided memory is invalid, or if the dimensions is invalid
+     */
+    copyFrom(memory src, vec<uint32_t> dimensions) generates (ErrorStatus status);
+};
diff --git a/neuralnetworks/1.3/IDevice.hal b/neuralnetworks/1.3/IDevice.hal
index 1295d6a..9afd778 100644
--- a/neuralnetworks/1.3/IDevice.hal
+++ b/neuralnetworks/1.3/IDevice.hal
@@ -22,6 +22,12 @@
 import @1.2::DeviceType;
 import @1.2::Extension;
 import @1.2::IDevice;
+import BufferDesc;
+import BufferRole;
+import Capabilities;
+import Model;
+import IBuffer;
+import IPreparedModel;
 import IPreparedModelCallback;
 
 /**
@@ -247,4 +253,61 @@
                               uint8_t[Constant:BYTE_SIZE_OF_CACHE_TOKEN] token,
                               IPreparedModelCallback callback)
             generates (ErrorStatus status);
+
+    /**
+     * Allocates a driver-managed buffer with the properties specified by the buffer descriptor
+     * as well as the input and output roles.
+     *
+     * The allocate function must verify its inputs are correct. If there is an error, or if a
+     * certain role or property is not supported by the driver, the allocate
+     * function must return with an appropriate ErrorStatus, a nullptr as the IBuffer, and 0 as the
+     * buffer token. If the allocation is successful, this method must return with ErrorStatus::NONE
+     * and the produced IBuffer with a positive token identifying the allocated buffer. A successful
+     * allocation must accommodate all of the specified roles and buffer properties.
+     *
+     * The buffer is allocated to an uninitialized state. An uninitialized buffer may only be used
+     * in ways that are specified by outputRoles. A buffer is initialized after it is used as an
+     * output in a successful execution, or after a successful invocation of IBuffer::copyFrom on
+     * the buffer. An initialized buffer may be used according to all roles specified in inputRoles
+     * and outputRoles. A buffer will return to the uninitialized state if it is used as an output
+     * in a failed execution, or after a failed invocation of IBuffer::copyFrom on the buffer.
+     *
+     * The dimensions of the buffer can be deduced from the buffer descriptor as well as the
+     * dimensions of the corresponding model operands of the input and output roles. The dimensions
+     * or rank of the buffer may be unknown at this stage. As such, some driver services may only
+     * create a placeholder and defer the actual allocation until execution time. Note that the
+     * same buffer may be used for different shapes of outputs on different executions. When the
+     * buffer is used as an input, the input shape must be the same as the output shape from the
+     * last execution using this buffer as an output.
+     *
+     * The driver must apply proper validatation upon every usage of the buffer, and must fail the
+     * execution immediately if the usage is illegal.
+     *
+     * @param desc A buffer descriptor specifying the properties of the buffer to allocate.
+     * @param preparedModels A vector of IPreparedModel objects. Must only contain IPreparedModel
+     *     objects from the same IDevice as this method is being invoked on.
+     * @param inputRoles A vector of roles with each specifying an input to a prepared model.
+     * @param outputRoles A vector of roles with each specifying an output to a prepared model.
+     *     Each role specified in inputRoles and outputRoles must be unique. The corresponding
+     *     model operands of the roles must have the same OperandType, scale, zero point, and
+     *     ExtraParams. The dimensions of the operands and the dimensions specified in the buffer
+     *     descriptor must be compatible with each other. Two dimensions are incompatible if there
+     *     is at least one axis that is fully specified in both but has different values.
+     * @return status Error status of the buffer allocation. Must be:
+     *     - NONE if successful
+     *     - DEVICE_UNAVAILABLE if driver is offline or busy
+     *     - GENERAL_FAILURE if a certain buffer property or a certain role is not supported,
+     *       or if there is an unspecified error
+     *     - INVALID_ARGUMENT if one of the input arguments is invalid
+     * @return buffer The allocated IBuffer object. If the buffer was unable to be allocated
+     *     due to an error, nullptr must be returned.
+     * @return token A positive token identifying the allocated buffer. The same token will be
+     *     provided when referencing the buffer as one of the memory pools in the request of an
+     *     execution. The token must not collide with the tokens of other IBuffer objects that are
+     *     currently alive in the same driver service. If the buffer was unable to be allocated
+     *     due to an error, the token must be 0.
+     */
+    allocate(BufferDesc desc, vec<IPreparedModel> preparedModels, vec<BufferRole> inputRoles,
+             vec<BufferRole> outputRoles)
+            generates (ErrorStatus status, IBuffer buffer, int32_t token);
 };
diff --git a/neuralnetworks/1.3/IPreparedModel.hal b/neuralnetworks/1.3/IPreparedModel.hal
index 7aea416..00adc1f 100644
--- a/neuralnetworks/1.3/IPreparedModel.hal
+++ b/neuralnetworks/1.3/IPreparedModel.hal
@@ -17,12 +17,12 @@
 package android.hardware.neuralnetworks@1.3;
 
 import @1.0::ErrorStatus;
-import @1.0::Request;
 import @1.2::IExecutionCallback;
 import @1.2::IPreparedModel;
 import @1.2::MeasureTiming;
 import @1.2::OutputShape;
 import @1.2::Timing;
+import Request;
 
 /**
  * IPreparedModel describes a model that has been prepared for execution and
@@ -33,7 +33,8 @@
      * Launches an asynchronous execution on a prepared model.
      *
      * The execution is performed asynchronously with respect to the caller.
-     * execute_1_3 must verify the inputs to the function are correct. If there is
+     * execute_1_3 must verify the inputs to the function are correct, and the usages
+     * of memory pools allocated by IDevice::allocate are valid. If there is
      * an error, execute_1_3 must immediately invoke the callback with the
      * appropriate ErrorStatus value, then return with the same ErrorStatus. If
      * the inputs to the function are valid and there is no error, execute_1_3 must
@@ -95,7 +96,8 @@
      *
      * The execution is performed synchronously with respect to the caller.
      * executeSynchronously_1_3 must verify the inputs to the function are
-     * correct. If there is an error, executeSynchronously_1_3 must immediately
+     * correct, and the usages of memory pools allocated by IDevice::allocate
+     * are valid. If there is an error, executeSynchronously_1_3 must immediately
      * return with the appropriate ErrorStatus value. If the inputs to the
      * function are valid and there is no error, executeSynchronously_1_3 must
      * perform the execution, and must not return until the execution is
diff --git a/neuralnetworks/1.3/types.hal b/neuralnetworks/1.3/types.hal
index 62c5833..6c8fe43 100644
--- a/neuralnetworks/1.3/types.hal
+++ b/neuralnetworks/1.3/types.hal
@@ -19,6 +19,7 @@
 import @1.0::DataLocation;
 import @1.0::OperandLifeTime;
 import @1.0::PerformanceInfo;
+import @1.0::RequestArgument;
 import @1.2::OperandType;
 import @1.2::OperationType;
 import @1.2::SymmPerChannelQuantParams;
@@ -5205,3 +5206,92 @@
         LOW_BITS_TYPE = 16,
     };
 };
+
+/**
+ * A buffer descriptor. Describes the properties of a buffer.
+ */
+struct BufferDesc {
+    /**
+     * Dimensions of the buffer. May have unknown dimensions or rank. A buffer with some number
+     * of unspecified dimensions is represented by setting each unspecified dimension to 0. A
+     * buffer with unspecified rank is represented by providing an empty dimensions vector.
+     */
+    vec<uint32_t> dimensions;
+};
+
+/**
+ * Describes a role of an input or output to a prepared model.
+ */
+struct BufferRole {
+    /**
+     * The index of the IPreparedModel within the "preparedModel" argument passed in
+     * IDevice::allocate.
+     */
+    uint32_t modelIndex;
+
+    /**
+     * The index of the input or output operand.
+     */
+    uint32_t ioIndex;
+
+    /**
+     * A floating-point value within the range (0.0, 1.0]. Describes how likely the
+     * buffer is to be used in the specified role. This is provided as a hint to
+     * optimize the case when multiple roles prefer different buffer locations or data
+     * layouts.
+     */
+    float frequency;
+};
+
+/**
+ * Inputs to be sent to and outputs to be retrieved from a prepared model.
+ *
+ * A Request serves two primary tasks:
+ * 1) Provides the input and output data to be used when executing the model.
+ * 2) Specifies any updates to the input operand metadata that were left
+ *    unspecified at model preparation time.
+ *
+ * An output must not overlap with any other output, with an input, or
+ * with an operand of lifetime CONSTANT_REFERENCE.
+ */
+struct Request {
+    /**
+     * Input data and information to be used in the execution of a prepared
+     * model.
+     *
+     * The index of the input corresponds to the index in Model.inputIndexes.
+     *   E.g., input[i] corresponds to Model.inputIndexes[i].
+     */
+    vec<RequestArgument> inputs;
+
+    /**
+     * Output data and information to be used in the execution of a prepared
+     * model.
+     *
+     * The index of the output corresponds to the index in Model.outputIndexes.
+     *   E.g., output[i] corresponds to Model.outputIndexes[i].
+     */
+    vec<RequestArgument> outputs;
+
+    /**
+     * A memory pool.
+     */
+    safe_union MemoryPool {
+        /**
+         * Specifies a client-managed shared memory pool.
+         */
+        memory hidlMemory;
+
+        /**
+         * Specifies a driver-managed buffer. It is the token returned from IDevice::allocate,
+         * and is specific to the IDevice object.
+         */
+        int32_t token;
+    };
+
+    /**
+     * A collection of memory pools containing operand data for both the
+     * inputs and the outputs to a model.
+     */
+    vec<MemoryPool> pools;
+};
diff --git a/neuralnetworks/1.3/types.t b/neuralnetworks/1.3/types.t
index 0d20d06..b1c72a9 100644
--- a/neuralnetworks/1.3/types.t
+++ b/neuralnetworks/1.3/types.t
@@ -21,6 +21,7 @@
 import @1.0::DataLocation;
 import @1.0::OperandLifeTime;
 import @1.0::PerformanceInfo;
+import @1.0::RequestArgument;
 import @1.2::OperandType;
 import @1.2::OperationType;
 import @1.2::SymmPerChannelQuantParams;
@@ -389,3 +390,92 @@
         LOW_BITS_TYPE = 16,
     };
 };
+
+/**
+ * A buffer descriptor. Describes the properties of a buffer.
+ */
+struct BufferDesc {
+    /**
+     * Dimensions of the buffer. May have unknown dimensions or rank. A buffer with some number
+     * of unspecified dimensions is represented by setting each unspecified dimension to 0. A
+     * buffer with unspecified rank is represented by providing an empty dimensions vector.
+     */
+    vec<uint32_t> dimensions;
+};
+
+/**
+ * Describes a role of an input or output to a prepared model.
+ */
+struct BufferRole {
+    /**
+     * The index of the IPreparedModel within the "preparedModel" argument passed in
+     * IDevice::allocate.
+     */
+    uint32_t modelIndex;
+
+    /**
+     * The index of the input or output operand.
+     */
+    uint32_t ioIndex;
+
+    /**
+     * A floating-point value within the range (0.0, 1.0]. Describes how likely the
+     * buffer is to be used in the specified role. This is provided as a hint to
+     * optimize the case when multiple roles prefer different buffer locations or data
+     * layouts.
+     */
+    float frequency;
+};
+
+/**
+ * Inputs to be sent to and outputs to be retrieved from a prepared model.
+ *
+ * A Request serves two primary tasks:
+ * 1) Provides the input and output data to be used when executing the model.
+ * 2) Specifies any updates to the input operand metadata that were left
+ *    unspecified at model preparation time.
+ *
+ * An output must not overlap with any other output, with an input, or
+ * with an operand of lifetime CONSTANT_REFERENCE.
+ */
+struct Request {
+    /**
+     * Input data and information to be used in the execution of a prepared
+     * model.
+     *
+     * The index of the input corresponds to the index in Model.inputIndexes.
+     *   E.g., input[i] corresponds to Model.inputIndexes[i].
+     */
+    vec<RequestArgument> inputs;
+
+    /**
+     * Output data and information to be used in the execution of a prepared
+     * model.
+     *
+     * The index of the output corresponds to the index in Model.outputIndexes.
+     *   E.g., output[i] corresponds to Model.outputIndexes[i].
+     */
+    vec<RequestArgument> outputs;
+
+    /**
+     * A memory pool.
+     */
+    safe_union MemoryPool {
+        /**
+         * Specifies a client-managed shared memory pool.
+         */
+        memory hidlMemory;
+
+        /**
+         * Specifies a driver-managed buffer. It is the token returned from IDevice::allocate,
+         * and is specific to the IDevice object.
+         */
+        int32_t token;
+    };
+
+    /**
+     * A collection of memory pools containing operand data for both the
+     * inputs and the outputs to a model.
+     */
+    vec<MemoryPool> pools;
+};
diff --git a/neuralnetworks/1.3/vts/functional/GeneratedTestHarness.cpp b/neuralnetworks/1.3/vts/functional/GeneratedTestHarness.cpp
index eced063..09ccc9a 100644
--- a/neuralnetworks/1.3/vts/functional/GeneratedTestHarness.cpp
+++ b/neuralnetworks/1.3/vts/functional/GeneratedTestHarness.cpp
@@ -60,7 +60,6 @@
 using V1_0::DataLocation;
 using V1_0::ErrorStatus;
 using V1_0::OperandLifeTime;
-using V1_0::Request;
 using V1_1::ExecutionPreference;
 using V1_2::Constant;
 using V1_2::MeasureTiming;
@@ -192,7 +191,7 @@
     return byteSize > 1u;
 }
 
-static void makeOutputInsufficientSize(uint32_t outputIndex, Request* request) {
+static void makeOutputInsufficientSize(uint32_t outputIndex, V1_0::Request* request) {
     auto& length = request->outputs[outputIndex].location.length;
     ASSERT_GT(length, 1u);
     length -= 1u;
@@ -245,10 +244,11 @@
         return;
     }
 
-    Request request = createRequest(testModel);
+    V1_0::Request request10 = createRequest(testModel);
     if (testConfig.outputType == OutputType::INSUFFICIENT) {
-        makeOutputInsufficientSize(/*outputIndex=*/0, &request);
+        makeOutputInsufficientSize(/*outputIndex=*/0, &request10);
     }
+    Request request = nn::convertToV1_3(request10);
 
     ErrorStatus executionStatus;
     hidl_vec<OutputShape> outputShapes;
@@ -284,6 +284,8 @@
             break;
         }
         case Executor::BURST: {
+            // TODO(butlermichael): Check if we need to test burst in V1_3 if the interface remains
+            //                      V1_2.
             SCOPED_TRACE("burst");
 
             // create burst
@@ -292,15 +294,15 @@
             ASSERT_NE(nullptr, controller.get());
 
             // create memory keys
-            std::vector<intptr_t> keys(request.pools.size());
+            std::vector<intptr_t> keys(request10.pools.size());
             for (size_t i = 0; i < keys.size(); ++i) {
-                keys[i] = reinterpret_cast<intptr_t>(&request.pools[i]);
+                keys[i] = reinterpret_cast<intptr_t>(&request10.pools[i]);
             }
 
             // execute burst
             int n;
             std::tie(n, outputShapes, timing, std::ignore) =
-                    controller->compute(request, testConfig.measureTiming, keys);
+                    controller->compute(request10, testConfig.measureTiming, keys);
             executionStatus = nn::convertResultCodeToErrorStatus(n);
 
             break;
@@ -361,7 +363,7 @@
     }
 
     // Retrieve execution results.
-    const std::vector<TestBuffer> outputs = getOutputBuffers(request);
+    const std::vector<TestBuffer> outputs = getOutputBuffers(request10);
 
     // We want "close-enough" results.
     checkResults(testModel, outputs);
diff --git a/neuralnetworks/1.3/vts/functional/ValidateRequest.cpp b/neuralnetworks/1.3/vts/functional/ValidateRequest.cpp
index 8092d04..96dc589 100644
--- a/neuralnetworks/1.3/vts/functional/ValidateRequest.cpp
+++ b/neuralnetworks/1.3/vts/functional/ValidateRequest.cpp
@@ -28,7 +28,6 @@
 namespace android::hardware::neuralnetworks::V1_3::vts::functional {
 
 using V1_0::ErrorStatus;
-using V1_0::Request;
 using V1_2::MeasureTiming;
 using V1_2::OutputShape;
 using V1_2::Timing;
@@ -93,9 +92,13 @@
     }
 
     // burst
+    // TODO(butlermichael): Check if we need to test burst in V1_3 if the interface remains V1_2.
     {
         SCOPED_TRACE(message + " [burst]");
 
+        ASSERT_TRUE(nn::compliantWithV1_0(request));
+        V1_0::Request request10 = nn::convertToV1_0(request);
+
         // create burst
         std::shared_ptr<::android::nn::ExecutionBurstController> burst =
                 android::nn::ExecutionBurstController::create(preparedModel,
@@ -103,13 +106,13 @@
         ASSERT_NE(nullptr, burst.get());
 
         // create memory keys
-        std::vector<intptr_t> keys(request.pools.size());
+        std::vector<intptr_t> keys(request10.pools.size());
         for (size_t i = 0; i < keys.size(); ++i) {
-            keys[i] = reinterpret_cast<intptr_t>(&request.pools[i]);
+            keys[i] = reinterpret_cast<intptr_t>(&request10.pools[i]);
         }
 
         // execute and verify
-        const auto [n, outputShapes, timing, fallback] = burst->compute(request, measure, keys);
+        const auto [n, outputShapes, timing, fallback] = burst->compute(request10, measure, keys);
         const ErrorStatus status = nn::convertResultCodeToErrorStatus(n);
         EXPECT_EQ(ErrorStatus::INVALID_ARGUMENT, status);
         EXPECT_EQ(outputShapes.size(), 0);
@@ -117,7 +120,7 @@
         EXPECT_FALSE(fallback);
 
         // additional burst testing
-        if (request.pools.size() > 0) {
+        if (request10.pools.size() > 0) {
             // valid free
             burst->freeMemory(keys.front());
 
diff --git a/neuralnetworks/1.3/vts/functional/VtsHalNeuralnetworks.cpp b/neuralnetworks/1.3/vts/functional/VtsHalNeuralnetworks.cpp
index 92d8fa7..1140b68 100644
--- a/neuralnetworks/1.3/vts/functional/VtsHalNeuralnetworks.cpp
+++ b/neuralnetworks/1.3/vts/functional/VtsHalNeuralnetworks.cpp
@@ -25,6 +25,7 @@
 #include "1.3/Callbacks.h"
 #include "GeneratedTestHarness.h"
 #include "TestHarness.h"
+#include "Utils.h"
 
 namespace android::hardware::neuralnetworks::V1_3::vts::functional {
 
@@ -32,7 +33,6 @@
         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_0::Request;
 using V1_1::ExecutionPreference;
 
 // internal helper function
@@ -124,9 +124,9 @@
 // Forward declaration from ValidateModel.cpp
 void validateModel(const sp<IDevice>& device, const Model& model);
 // Forward declaration from ValidateRequest.cpp
-void validateRequest(const sp<IPreparedModel>& preparedModel, const V1_0::Request& request);
+void validateRequest(const sp<IPreparedModel>& preparedModel, const Request& request);
 // Forward declaration from ValidateRequest.cpp
-void validateRequestFailure(const sp<IPreparedModel>& preparedModel, const V1_0::Request& request);
+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);
 
@@ -139,7 +139,11 @@
     if (preparedModel == nullptr) return;
 
     validateRequest(preparedModel, request);
-    validateBurst(preparedModel, request);
+
+    // TODO(butlermichael): Check if we need to test burst in V1_3 if the interface remains V1_2.
+    ASSERT_TRUE(nn::compliantWithV1_0(request));
+    V1_0::Request request10 = nn::convertToV1_0(request);
+    validateBurst(preparedModel, request10);
 }
 
 void validateFailure(const sp<IDevice>& device, const Model& model, const Request& request) {
@@ -157,7 +161,7 @@
 
 TEST_P(ValidationTest, Test) {
     const Model model = createModel(kTestModel);
-    const Request request = createRequest(kTestModel);
+    const Request request = nn::convertToV1_3(createRequest(kTestModel));
     if (kTestModel.expectFailure) {
         validateFailure(kDevice, model, request);
     } else {