Autonomic Computing aims at realizing computing systems that are able to adapt themselves, but the engineering of such systems in the large is rather a challenging task. It is hard to find an appropriate model that controls the adaptation itself and several loops are likely to be coordinated to avoid unexpected and harmful behaviors. This paper presents an approach and a runtime support to architecture self-adaptive systems, in which each part of the feedback control loop is uniformly and explicitly designed as a first-class adaptive element. Making these elements explicit allows the architect to reason about system modeling and to reuse them. Code generation from the architecture model avoids painful details of low-level system implementation.