Assuming you're running on hotspot and as long as the lifetime of the slice is very shortlived, e.g. immediately used in the method creating it or by its caller, then escape analysis should be able to eliminate that allocation.
That is a JVM optimization, so there's no guarantee that it happens, but it generally is good enough to not worry about those things.
Also, young GCs are very efficient. The cost of such short-lived objects is very low, even if EA does not kick in.
Also, you should avoid premature optimizations. Worry about such things once you measured performance and figured out where the actual bottlenecks are.