Dhaval,
You’re in classic parallel convoy territory here (links at the end of the reply) – a parallel convoy allows you to wait for two correlated messages before processing starts on those messages (In this case, your ‘unique-no’ would be the value that is used to correlate the two messages).
In the scenario above, either the item or header file can be received first, which will instanciate the orchestration. It will then wait until the other message is received before continuing with the orchestration. The only slight issue I see here is where you will take the ‘unique-no’ from. If it is also in the message, you will just need to promote the element into the message context (not a distinguished field) and then use that as the correlation value. If it is only in the filename, you will need to write a receive pipeline decode component to take the unique-no out of the filename and promote it.
HTH, Nick.