Native memory leak with frequent index delete/create

As part of some flow, we used to create and delete an index just to do smoke test kind of regularly. Over months, we saw native memory is growing very badly. And finally, it touches max limits of OS memory and crashes.
Basically, we see ~1MB growth in native memory for every index create/delete. I just kept a loop for 5K and see native growth of 4.5GB!!!!

By the way, native memory memory = (Total process memory usage - Heap Size)

And even after leaving cluster free for some time, its is not going down.
Can anyway please suggest whats the issue? Any ES specific setting to limit this growth.

We monitored the pmap output to see what exactly growing. We see that some 64MB blocks are getting created and used more and more. At the start it was 33 blocks and after 5K loop, it went to 110 blocks. pmap doesn't give any details on what these blocks are.