I wouldn't recommend fitting a piecewise model to the complete data set. Rather, divide the data into a linear and an exponential part and fit each one separately. You can use polyfit() to fit the linear part. For the exponential part, you could use nlinfit as long as there aren't any bound constraints that you would like to put on the exponential parameters. If you wish to apply bounds, I would recommend either fit() or lsqcurvefit(), depending on which Toolboxes you have.
Best Answer