Yes, there's a global accumulator, but just because there's a global utxo set. You can have purpose specific accumulators if you want.
However, all acculators I know need to update the witness on both add and del. I need to look at the paper again, but there's a provable tradeoff between update rate and size. You can't get a succinct acculator that also doesn't update the witness, so if you can't assume witness updates, IFAIK acculators won't do. (I might be wrong though, I'm not a cryptographer)