Pervasive systems are often made out of distributed software components that run on different computational units (appliances, sensing and actuating devices, computers). Such components are often developed, maintained, and even operated by different parties. Applications are increasingly built by dynamically discovering and composing such components in a situation-aware manner. By this we mean that applications follow some strategies to self-organize themselves to adapt their behavior depending on the changing situation in which they operate, for example the physical environment. They may also evolve autonomously in response to changing requirements. Software architectures are considered a well-suited abstraction to achieve situational adaptation. In this paper, we review some existing architectural approaches to self-adaptation and propose a high-level meta-model for architectures that supports dynamic adaptation. The meta-model is then instantiated in a specific ambient computing case study, which is used to illustrate its applicability.