Hello shrinand ,
I am not very familiar with deepsort, so apologies for potentially stupid questions. Is the blob itself for yolov4+deepsort, or are those 2 separate blobs? If two separate, you could run eg. yolov4 on the OAK, and run deepsort on the host machine (RPI) - if it's only one, you could only run it completely on the OAK, which might have performance bottlenecks.
On how to implement - I would look at efficientDet demo. So you could just change previewSize, blob path and host-side decoding and have it working!
Thanks, Erik