Access to blending units?


In a number of problems I have been working on, some sort of access to blending units(eg adding or multiplying a value when writing to an image) would be very helpful. However, unlike texture units, I have not seen any provision for this in compute APIs. Is this something I have missed(perhaps in CUDA) or is it implimented in an indirect a way for graphics drivers which precludes the functionality from being exposed in say a OpenCL extension or CUDA feature?