You can start by approximating the ray distribution of the source file with a simple model in sequential mode. The spatial ray distribution can be modeled by a set of object height fields that are distributed according to the shape of the source, e.g. rectangular for an LED. The angular ray distribution can be modelled by setting the aperture type to “Object Cone Angle” and choosing an appropriate aperture value. 

Of course the actual ray distribution from the source file is probably much more complex than that. You could at first find a lens system in sequential mode that fulfills your needs within the range of the simplified model. Next, this lens system could be fine-tuned with the actual ray distribution in the non-sequential mode.

HOW CAN I OPTIMIZE THE LENSES IN NON-SEQUENTIAL MODE WHEN I HAVE DESIGNED THEM FOR A SIMPLIED SOURCE IN SEQUENTIAL MODE?

The transfer of the system from sequential to non-sequential mode can be done with the “Convert to NSC group” tool. In NSC mode, you can check with the source file whether your lens system achieves the desired result in terms of collimating the source. If this is not the case, you can further try to optimize the system, for example by minimizing the angular spread. Since you start with a system that should already be somewhere near a local optimum, the non-sequential optimization should not take too long.

Thank you. Can you provide me link of any available article for optimization in non-sequential mode for collimation?

The following article from the knowledge base mentions optimization for collimation:
How to optimize non-sequential optical systems – Knowledgebase (zemax.com)