GPU@SAT framework is capable of taking as input AI models described using the most common AI frameworks, such as TensorFlow, PyTorch, and ONNX.  However, the generated code is provided as a group of kernels that have to be run on the IP-Core in a certain order. The development of the orchestrator code to prepare and execute these kernels is left to end users. This can cause difficulties in the development and testing of the applications, due to the use of multiple programming languages like C, C++, and OpenCL. The objective of the sub-activity is to develop and test an enhanced GPU@SAT framework that generates a modular skeleton application. The skeleton shall contain the memory management and kernel execution as foreseen by the AI models. Furthermore, if one or multiple layers inside the model are not supported, the enhanced GPU@SAT framework shall generate one or multiple placeholder functions for implementing these layers directly in OpenCL.