What is the best way to code a scientific program? [on hold]

I am trying to code a scientific program, on C++, that performs some calculations on arrays, matrices... and I want it to be performant. I see that C++ allows OOP methods but it is said that this highly reduces the performances of the program.

According to you, shoud I use OOP (this may make my program more organized), what about the try an catch blocks, templates... Do all these things reduce the performance of the program? What are the best practices...

Thank you in advance.