it is not true that you have to do that in order to type a non-linear model into dynare. It is possible to enter equations in both ways you described.
Dynare actually calculates a linearization of the model-equations that you type in. What follows is that if you give it the equation as Y = C + I and when you then use stoch_simul it calculates absolute deviations from the non-stochastic steady state. The ‘trick’ with exp(Y) = exp( C) + exp(I) is that dynare then does a log-linearization for you and you will receive percentage deviations form steady state after you do stochastic simulations.
This is the reason many enter the model-equations that way, as they rather be interpreting percentage deviations.