Why do you want to stretch the values between -1 and 1? Aren't they allready inside it? If they aren't then there is a problem with the original data (i.e. negative values).
With good data set and correctly calculated NDVI you'll always get a ratio between -1 and 1 no matter what the units you put in are (DN, radiance or reflectance). Reflectance with atleast atmospheric correction should give the most reliable results.