Machine Learning is making rapid progress in a variety of applications. It is highly likely to be used in safety-related and possibly safety-critical systems. As a logical next step to work presented at the Safety-Critical Systems Symposium 2019 on developing a safety argument structure for an autonomous system that uses machine learning, this paper focuses on generating the underpinning safety evidence. This is achieved through the representation of the machine-learnt software development life-cycle as a model which articulates constituent artefacts, information flow and transformations. This life-cycle model is then used to facilitate the systematic identification of the potential for the introduction of hazardous errors during development. Product, process and goal based control measures are proposed to reduce and manage these potential errors. The feasibility and practicality of implementing these control measures and generating associated safety evidence is also discussed.