Cyber physical systems (CPSs) by definition involve design constraints addressing the computation and communication necessary to control physical systems. These systems have been modeled using domain specific modeling languages, but some limitations exist in the continued application of such a modeling approach to more complex, or safety-critical, systems. Specifically, it is well known how to formulate constraints in a domain-specific modeling language in order to prevent users from building invalid structures, but existing constraint-based techniques do not take into consideration design requirements that may require analysis in the physical domain (i.e. dynamic constraints). Those analysis results, when interpreted by a domain expert, can inform changes to the model: unfortunately, this "by hand" process does not scale. This paper presents an approach to automate the process of evolving models based on dynamic constraints that are not structurally enforceable into the modeling of CPSs. This new methodology-called dynamic constraint feedback (DCF)-is described herein and demonstrated with specific examples derived from the domain of data adaptable reconfigurable embedded systems (DARES). In DCF, expert blocks are integrated with a modeling language to perform dynamic constraint analysis on system models. The results from these analyses are then used to generate model transformations that can be applied to the source models.