diff --git a/README.md b/README.md
index 9058b1e..b8ebb7c 100644
--- a/README.md
+++ b/README.md
@@ -134,6 +134,14 @@ The `inference_pkg_launch.py`, included in this package, provides an example dem
 |`load_model`|`LoadModelSrv`|Service that is responsible for setting pre-processing algorithm and inference tasks for the specific type of model loaded.|
 |`inference_state`|`InferenceStateSrv`|Service that is responsible for starting and stopping inference tasks.|
 
+
+### Parameters
+
+| Parameter name   | Description  |
+| ---------------- |  ----------- |
+| `device` | String that is either `CPU`, `GPU` or `MYRIAD`. Default is `CPU`. `MYRIAD` is the Intel Compute Stick 2. |
+
+
 ## Resources
 
 * [Getting started with AWS DeepRacer OpenSource](https://github.com/aws-deepracer/aws-deepracer-launcher/blob/main/getting-started.md)
diff --git a/inference_pkg/include/inference_pkg/inference_base.hpp b/inference_pkg/include/inference_pkg/inference_base.hpp
index 0dd4208..6cddcb1 100644
--- a/inference_pkg/include/inference_pkg/inference_base.hpp
+++ b/inference_pkg/include/inference_pkg/inference_base.hpp
@@ -32,8 +32,10 @@ namespace InferTask {
         /// @returns True if model loaded successfully, false otherwise
         /// @param artifactPath Path to the model artifact.
         /// @param imgProcess Pointer to the image processing algorithm
+        /// @param device Reference to the compute device (CPU, GPU, MYRIAD)
         virtual bool loadModel(const char* artifactPath,
-                               std::shared_ptr<ImgProcessBase> imgProcess) = 0;
+                               std::shared_ptr<ImgProcessBase> imgProcess,
+                               std::string device) = 0;
         /// Starts the inference task until stopped.
         virtual void startInference() = 0;
         /// Stops the inference task if running.
diff --git a/inference_pkg/include/inference_pkg/intel_inference_eng.hpp b/inference_pkg/include/inference_pkg/intel_inference_eng.hpp
index 47738f0..4295950 100644
--- a/inference_pkg/include/inference_pkg/intel_inference_eng.hpp
+++ b/inference_pkg/include/inference_pkg/intel_inference_eng.hpp
@@ -34,7 +34,8 @@ namespace IntelInferenceEngine {
         RLInferenceModel(std::shared_ptr<rclcpp::Node> inferenceNodePtr, const std::string &sensorSubName);
         virtual ~RLInferenceModel();
         virtual bool loadModel(const char* artifactPath,
-                               std::shared_ptr<InferTask::ImgProcessBase> imgProcess) override;
+                               std::shared_ptr<InferTask::ImgProcessBase> imgProcess,
+                               std::string device) override;
         virtual void startInference() override;
         virtual void stopInference() override;
         /// Callback method to retrieve sensor data.
diff --git a/inference_pkg/src/inference_node.cpp b/inference_pkg/src/inference_node.cpp
index 56e71a5..12628fd 100644
--- a/inference_pkg/src/inference_node.cpp
+++ b/inference_pkg/src/inference_node.cpp
@@ -41,11 +41,18 @@ namespace InferTask {
     /// Class that will manage the inference task. In particular it will start and stop the
     /// inference tasks and feed the inference task the sensor data.
     /// @param nodeName Reference to the string containing name of the node.
+    /// @param device Reference to the compute device (CPU, GPU, MYRIAD)
     public:
         InferenceNodeMgr(const std::string & nodeName)
-          : Node(nodeName)
+          : Node(nodeName),
+          deviceName_("CPU")
         {
             RCLCPP_INFO(this->get_logger(), "%s started", nodeName.c_str());
+
+            this->declare_parameter<std::string>("device", deviceName_);
+            // Device name; OpenVINO supports CPU, GPU and MYRIAD
+            deviceName_ = this->get_parameter("device").as_string();
+
             loadModelServiceCbGrp_ = this->create_callback_group(rclcpp::callback_group::CallbackGroupType::MutuallyExclusive);
             loadModelService_ = this->create_service<deepracer_interfaces_pkg::srv::LoadModelSrv>("load_model",
                                                                                                   std::bind(&InferTask::InferenceNodeMgr::LoadModelHdl,
@@ -129,7 +136,7 @@ namespace InferTask {
                         RCLCPP_ERROR(this->get_logger(), "Unknown inference task");
                         return;
                 }
-                itInferTask->second->loadModel(req->artifact_path.c_str(), itPreProcess->second);
+                itInferTask->second->loadModel(req->artifact_path.c_str(), itPreProcess->second, deviceName_);
                 res->error = 0;
             }
         }
@@ -149,6 +156,9 @@ namespace InferTask {
         /// List of available pre-processing algorithms.
         std::unordered_map<int, std::shared_ptr<ImgProcessBase>> preProcessList_;
         /// Reference to the node handler.
+
+        /// Compute device type.
+        std::string deviceName_;     
     };
 }
 
diff --git a/inference_pkg/src/intel_inference_eng.cpp b/inference_pkg/src/intel_inference_eng.cpp
index e9e2c40..6c65277 100644
--- a/inference_pkg/src/intel_inference_eng.cpp
+++ b/inference_pkg/src/intel_inference_eng.cpp
@@ -201,7 +201,8 @@ namespace IntelInferenceEngine {
     }
 
     bool RLInferenceModel::loadModel(const char* artifactPath,
-                            std::shared_ptr<InferTask::ImgProcessBase> imgProcess) {
+                            std::shared_ptr<InferTask::ImgProcessBase> imgProcess,
+                            std::string device) {
         if (doInference_) {
             RCLCPP_ERROR(inferenceNode->get_logger(), "Please stop inference prior to loading a model");
             return false;
@@ -214,7 +215,7 @@ namespace IntelInferenceEngine {
         imgProcess_ = imgProcess;
         // Load the model
         try {
-            inferRequest_ = setMultiHeadModel(artifactPath, "CPU", core_, inputNamesArr_,
+            inferRequest_ = setMultiHeadModel(artifactPath, device, core_, inputNamesArr_,
                                      outputName_, InferenceEngine::Precision::FP32,
                                      InferenceEngine::Precision::FP32, inferenceNode);
             for(size_t i = 0; i != inputNamesArr_.size(); ++i) {