What you described is similar in concept to Reservoir Computing. But those tend to have fixed reservoirs with learning occurring only on the readout stage.
You might also think about your “core cells” in terms of extended activation via Metabotropic Receptors. That came up on the mailing list last year iirc.