Software of Unknown Provenance, SOUP, refers to a software component that is already developed and widely available from a 3rd party, and that has not been developed, to be integrated into a medical device. From regulatory perspective, SOUP software requires special considerations, as the developers' obligations related to design and implementation are not applied to it. In this paper, we consider the implications of extending the concept of SOUP to machine learning (ML) models. As the contribution, we propose practical means to manage the added complexity of 3rd party ML models in regulated development.