One approach to increasing the sustainability of critical systems is to fortify them with cyber infrastructure that monitors the system, enables early diagnosis of faults, and provides decision support that facilitates greater efficacy. Modeling and analysis of the resulting cyber-physical systems is a significant challenge, as the physical and cyber infrastructures can be very different in time scales, complexity, and architecture. Model composition is a potential solution. Metamodeling seeks to facilitate model composition by providing methods for composing models of different types, including performance and dependability models, as well as methods for transforming one type of model to another. This paper describes the application of metamodeling to cyber-physical systems. Our proposed approach to model transformation is based on abstract interpretation, a program analysis technique. Models exist for disparate attributes of cyber-physical systems, but these models are typically domain-specific. We seek to map models from one physical domain to another, or to extract a model of one system attribute from a model of another attribute. This ability would considerably increase the utility of the existing body of knowledge on modeling of cyber-physicals systems.