What is the limit for number of classes in a multi-class classification using OpenCV

edit

I am currently working on a project where we have to build a model which can detect and recognize faces in a live-stream. I am using Python, OpenCV, and Haar Cascade Classifiers(HCC) for this project.

i have stumbled upon following questions while doing this:

1. How many different number of people(or classes) can OpenCV/HCC accurately detect if memory isn't a problem?

2. If I want to detect up to 10,000(or even a bigger number) different people, what are the difficulties I will come across using this technology which is fairly simple compared to other Deep Learning Models like ResNet, etc.?

3. If I cannot store images on my local machine when I scale up the number of ppl to be detected, what might be the best solution?

Add additional inputs if you have any. Would be of great help. Thank you.