The objective behind header compression for an IP packet is just to save radio resources or can say optimize use of radio resources.
For point to point bearer, it is performed at PDCP layer at eNodeB by using ROHC. But in case of multicast/broadcast data transmission , BM-SC does.
While establishing an MBMS session, BM-SC server sends MBMS header compression flag to MBMS-GW and GW forward the same information to MME. So this way, MME get to know that whether header compression is being used or not for that MBMS bearer.
Hopefully, I answered your query.