Is it reasonable to assume these tasks will be in the scheduler for a "long" period of time? If so, perhaps the best idea is the simplest possible in-kernel schedule builder coupled with an optional user-space schedule builder. Using a more complicated scheduler (and admission decider) adds a large initial cost, but that *may* be acceptable if the initial cost is much smaller than the total span of time the task will be in the scheduler. Statically analyzed systems and those with only one or two such tasks may be content with the in-kernel simple solution.