What is the total table size? If the whole table is 20M rows, you're probably better off breaking replication, doing the delete, and then taking a new snapshot to start with of the table with only 5M rows now. If the table has 2B rows, obviously that would not be a good idea :-D.

If at all possible, delete the rows in batches and give replication time to get "caught up". For example, delete 40,000 rows, wait a bit, delete 40,000 more, etc.. How long to wait depends on your drive speeds, speed of the network to your replicated data, etc..