Overlay networks support a wide range of peer-to-peer media streaming applications on the Internet. The user experience of such applications is affected by the churn resilience of the system. When peers disconnect from the system, streamed data may be delayed or lost due to missing links in the overlay topology. In this paper, we explore a proactive strategy to create churn-aware overlay networks that reduce the potential of disruptions caused by churn events. We describe CHAMS, a middleware for constructing overlay networks that mitigates the impact of churn. CHAMS uses a "hybrid" approach-it implicitly defines an overlay topology using a gossip-style mechanism, while taking the reliability of peers into account. Unlike systems for overlay construction, CHAMS supports a variety of topologies used in media streaming systems, such as trees, multi-trees and forests. We evaluate CHAMS with different topologies and show that it reduces the impact of churn, while imposing only low computational and message overheads.