I go back into my previous message: use GEKO and tune it to be as similar as your modified k-epsilon. Or define the LES part in the same way as your RANS part. No more solutions come to my mind.
Just out of curiosity: why do you need a modified k-epsilon and a modified blending function?