A key aspect in creating autonomous dynamic software systems is the possibility of reasoning about properties of runtime variability and dynamic behavior, e.g. when and how to reconfigure the system. Currently these properties are often not made explicit in the software architecture. We argue that having explicit models for the different architectural elements is a necessity, as it enables separation of concerns both pre-deployment and postdeployment. In this paper we propose a reconfiguration process based on a number of explicit models. Using this process as a basis for architectural development facilitates the design and evolution of autonomous reconfiguration.