Could you clarify the question? Linear regression is linear in terms of the model which is of an additive form such as y = b0 + b1*X1 + ... bk*Xk + e.
So I'm not sure what you mean by a non-linear IV. Box-Cox is used to normalise and/or stabilise variance which may or may not make the variables suitable for a linear model. It is also imperfect and can obscure interpretation.
I would usually start with an appropriate linear model and then check the assumptions by looking at the residuals of the model - as most of the assumptions (normality, homogeneity of variance and independence) are about the population of errors the residuals are sampled from, not about the raw variables.
well, Mr Buguley i have some non linear independent variables and i need to put them in the linear model so i've thought about making a Box cox transformation and then generate the linear model
Sorry - I perhaps wasn't clear. In what sense are the IVs non-linear? A single variable can't be non-linear. You must be referring to some other property:
e.g.,
curvilinear relationship with the outcome (Y)
lack of normality
lack of normality of residuals when in a model predicting Y
lack of independence of residuals when in a model predicting Y
heterogeneity of variance of residuals when in a model predicting Y