If I have expressed my model in logs and not levels, for e.g. in the preamble I declare a variable y and in the model block I express it as exp(y) and I solve the model using an approximation of order 1. Is the interpretation of the y axis of my IRF a percentage deviation from SS?

I also have debt in my model, which I do not express as exp because it can have a negative SS value. So is the IRF interpretation for debt just a deviation from SS?

I have a question along the same lines. I estimated a small open economy model , and since the steady state is different from 0, I created new variables to have the deviation from steady state interpretation in the impulse responses. Is this procedure wrong, considering the previous answer ?

IRFs are always relative to a baseline value. If the deviation from this baseline is in logs, it has the interpretation of percentage deviations. For IRFs, there is no reason to recenter everything as you do. However, subtracting the mean is not wrong.