Embodiments of the invention enable fast context switching of application specific processors having functional units with an architecturally visible state. In example embodiments, a processor allocates memory space to store two process control blocks for two active tasks to be performed by the processor comprising one or more custom functional units having a respective processing state not accessible by the processor. A memory controller stores the processing state of the custom functional units currently running a first active task, in a first process control block, in response to a preemptive task switch requirement. The memory controller loads a second processing state of the custom functional units for a second active task, from a second process control block in the memory, in response to the preemptive task switch requirement. The processor may then perform the second active task, based on the second processing state loaded into the custom functional units.