This paper proposes a reconfigurable data flow (RDF) model - an architecture meta-model for dynamic reconfiguration with QoS management. The RDF model is an extension to the widely used data flow computing model. The RDF model supports a set of reconfiguration actions that can change its structure during execution without interrupting the flow of data. The influence of a reconfiguration to the system QoS is identified as logical influence and performance influence. Both design time and runtime support is provided to cope with the influence. The logical influence can be minimized through properly designing a reconfiguration plan under a set of reconfiguration constraints. The performance influence can be controlled by a reconfiguration scheduling mechanism during runtime. The experimental results illustrate that using the reconfiguration constraints and the reconfiguration scheduling mechanism could evidently reduce the influence of a reconfiguration to the system QoS.