In this paper, we study how to schedule a set of real-time tasks where each task implements a control law. These tasks share a limited computing resource. The set of tasks can switch on line from one given configuration to another one depending on the working modes of the global application. This means that some tasks may appear while other ones may be no longer activated and that the WCET of some tasks may be modified. We propose a scheduling architecture for the handling of such task instances. At each mode switching, the task handler consider the new task parameters; then it determines on line a (m,k)-constraint based scheduling strategy to apply to each task; this strategy aims to selectively discard task instances so that the schedulability of tasks is guaranteed and the overall control performance is maintained at a high level.