You know in cointegration theory, we should use variables that are I(1). In your case just make the first differences of your variables. Thus, they become I(1), and make the cointegration relationship among them.

Dear Said Jaouadi, should the variables be integrated of the same order or only I(1) ? In fact, I think, if they are integrated of the same order ( 1, 2, or ...), and there exist a linear combination that comes out with a stationary residual , thus they are integrated. Therefore, Dear Vipin, using the Impulse Responses Functions (IRF) can help you to see the way a shock to one variable is reflected immediately to the other variables and the behavior of the transmitted shock overtime. If you use the option I am suggesting you ( IRF), you can use your raw variables without making them stationary because, according to other authors, when we are interested in the way variables react to shocks that occur to other variables , we are not much interested in the values of estimates. Please Read Walter Enders, Applied Econometric time series, chapter V & VI. However, IRFs are not ordering invariant. So if you chose IRF , you will need to come up with strong theoretical reasons to support the ordering of variables in your SVAR.

heartfelt thanks to you both ..................

you should use Impulse Responses Functions (IRF) and Variance Decomposition analysis. If your variables mix of  order  '(0)' and  order(I) use ARDL model.