Hello Arun,
There are two major potential sources for the delay that I can think of offhand. One is that using subscribers on the topic involves copying the data in the message. For large amounts of data, like images, this can take considerable time. I would suggest ensuring that you are making use of enabled subsystems (enabling them with the IsNew output from the Subscribe block), to avoid passing along and copying data that is not actually from a new message. This also should cut down on time by avoiding processing non-new messages.
Another (usually larger) factor in slowing down processing of anything is visualization. In MATLAB, this typically comes from plotting data or outputting text to the Command Window. In Simulink, this includes using Scopes or viewing images. If your model is sensitive to the rate, I would highly suggest removing all visualization while the model is running. Instead, collect the output for later viewing.
Another possibility, if you are testing an algorithm, is to collect the camera data and save it, then process the file through your algorithm later. You can either save the data as a rosbag on your ROS machine or subscribe to the topic from within MATLAB, directly saving the data to a MAT file. If you are able to update to R2018b, you could even play back the rosbag within Simulink, and process the data with model time synced to the rosbag timestamps.
Hope this helps you to get things moving.
-Cam
