you can try nonlinear transformation of these variables, for instance using the Log transformation. Also you need to keep the level of integration common among all the variables, in other words, if you are going to take the first difference then do it for all the control variables and make them all I(1). You need also to know why are the variables are nonstationary because this will define the way of transforming them. For instance, if the variables have a deterministic trend, then adding time to your list of control variables will do the job without applying any further differencing. Good luck

Please refer to this question for other ideas https://www.researchgate.net/post/If_my_variables_havedifferent_order_of_integrationGDP_I1_Population_I2_how_can_I_check_the_cointegration_among_these_variables. Also if you end up with variables at different levels of integration, then you can use ARDL model, in your case of panel data GMM. 

I am very astonished to see the list of independent variables. female participation in parliament is in form if number or ratio , . what are other two variables not mentioned in question. wages are always linked with employment opportunities, which should be included as indepedent variable

Dear Sofia Anwar,

My other two variables are education level and FDI % of GDP. I added female participation in parliament, education level and female labour participation as control variables. should I exclude female participation in parliament variable? I was confused about it, I could not find many cross-country studies about the topic.Could you advice any other independent variable? Thank you so much for your comments and helps.

Dear Sherif Maher Hassan

Thank you so much for your helps.

I founded some articles about it and they suggested to look at cointegration between non-stationary variables. Is it correct way to solve tis problem? But I am a bit lost about   including the variables after cointegration test into the model and estimation?

Thank you so much for your helps.

As said, you try nonlinear transformation of the variables first, take their log. and see if they become stationary or not. if yes, your problem is solved and proceed with your model. This method is conditioned to the fact that you can use the absolute form of the variables, as it is not correct to log transform the variables in percentages. To learn more about the integration and unit root tests, refer to Gujarati and porter, (2008) book, chapter 21. 

Dear Hassan,

Thank you so much for your answers. I took all variables log but three of them are still non-stationary. Also I checked the cross-section dependency for all variables and they are cross-section dependent variable. As you said above, I am trying to run Panel ARDL model on my estimation but the cross-section dependency is a problem for my estimation? Thank you so much.

Gulsah Adam how did you solve your problem? If the data is still non-stationary after log transformation