A physical based framework for designing efficient and high-performing shape memory functional devices has been proposed. This framework integrates various important elements such as micromechanics of shape memory alloys (SMAs), phase transformation theory, continuum mechanics, and topology optimization, all of which are integrated with in a machine learning toolbox for functional design purposes. The continuum model foundation of the framework relies on micromechanical principles of phase transformation which are formulated in a thermodynamically consistent way. This model is implemented using finite element software and linked with a machine learning algorithms to create fast and efficient surrogate models. To evaluate the performance, the framework has been applied to the design of a shape memory aortic stent device through combined material and geometry optimization. Design specifications, material requirements, and geometrical constraints are extracted from relevant literature and technical specifications. As a result of this optimization process, a stent design with improved performance is achieved, with a 35% increase in stretching ability and a 60% reduction in maximum stress and fatigue strain, thereby reducing the risk of fatigue and mechanical failure while maintaining the required stretching capacity of the aortic tissue.