Combining cross-layering and autonomic paradigms is considered, until now, the best alternative for better Quality of Service (QoS) support in wireless networks. We adopt this combination in the design of a new architecture called Cross- Layer Engine (XLEngine) having the ability to make optimizations based on local and network wide knowledge. The unique mode to acquire this network knowledge is that every node notifies other nodes about its local metrics (called local view). To achieve this goal, we propose a new protocol called the Local View Management Protocol (LVMP). LVMP implements a selective rebroadcast technique to reduce the number of nodes participating in the local view dissemination process while assuring that this local view attains all nodes. We give a performance evaluation of LVMP and we study the impact of the overhead on the overall XLEngine performance.