The development of Non-Terrestrial Networks in modern communication standards like 5G includes a functional split of base station functionalities. This introduces the demand for high-performance processing on-board satellites. As a possible solution, this paper presents the architectural design for a high-performance processing platform as a demonstrator for Machine-Learning-based on-board baseband processing for 5G Non-Terrestrial Networks. A Versal AI Edge device is the centerpiece of the processing unit, providing specialized acceleration for ML algorithms. Additionally, a hardware platform to implement the functionality of the ground segment is presented.