In order to effectively program a novel hardware accelerator, the DCC group at OFFIS developed a new Dataflow Graph (DFG) language. This novel accelerator is specifically suited for the acceleration of machine learning algorithms. Thus, it is important to enable a simple entry point to the DFG language from common machine learning frameworks (e.g., Tensorflow). The ONNX format is an interesting format for this task since it is generally independent of the concrete machine learning framework. The task of this thesis is to provide a systematic translation process for parametric ONNX models to the DFG language, thereby providing interesting insights of acceleration possibilities for machine learning algorithms on novel cutting-edge hardware accelerators.