diff --git a/sklearn/_build_utils/__init__.py b/sklearn/_build_utils/__init__.py
index a2e69c6978efe..1d6c1eaf607a6 100644
--- a/sklearn/_build_utils/__init__.py
+++ b/sklearn/_build_utils/__init__.py
@@ -75,3 +75,30 @@ def maybe_cythonize_extensions(top_path, config):
             nthreads=n_jobs,
             compile_time_env={'SKLEARN_OPENMP_SUPPORTED': with_openmp},
             compiler_directives={'language_level': 3})
+
+
+def gen_from_templates(templates, top_path):
+    """Generate cython files from a list of templates"""
+    is_release = os.path.exists(os.path.join(top_path, 'PKG-INFO'))
+    # Files are already cythonized, nothing to do.
+    if is_release:
+        return
+
+    # Lazy import because cython is not a dependency when building from
+    # source distribution.
+    from Cython import Tempita # noqa
+
+    for template in templates:
+        outfile = template.replace('.tp', '')
+
+        # if the template is not updated, no need to output the cython file
+        if not (os.path.exists(outfile) and
+                os.stat(template).st_mtime < os.stat(outfile).st_mtime):
+
+            with open(template, "r") as f:
+                tmpl = f.read()
+
+            tmpl_ = Tempita.sub(tmpl)
+
+            with open(outfile, "w") as f:
+                f.write(tmpl_)
diff --git a/sklearn/linear_model/setup.py b/sklearn/linear_model/setup.py
index b7ef0a7aa39f5..121b449d673d0 100644
--- a/sklearn/linear_model/setup.py
+++ b/sklearn/linear_model/setup.py
@@ -1,7 +1,9 @@
 import os
-
 import numpy
 
+from sklearn._build_utils import gen_from_templates
+
+
 def configuration(parent_package='', top_path=None):
     from numpy.distutils.misc_util import Configuration
 
@@ -22,19 +24,8 @@ def configuration(parent_package='', top_path=None):
                          libraries=libraries)
 
     # generate sag_fast from template
-    sag_cython_file = 'sklearn/linear_model/_sag_fast.pyx.tp'
-    sag_file = sag_cython_file.replace('.tp', '')
-
-    if not (os.path.exists(sag_file) and
-            os.stat(sag_cython_file).st_mtime < os.stat(sag_file).st_mtime):
-
-        with open(sag_cython_file, "r") as f:
-            tmpl = f.read()
-        from Cython import Tempita # noqa
-        tmpl_ = Tempita.sub(tmpl)
-
-        with open(sag_file, "w") as f:
-            f.write(tmpl_)
+    templates = ['sklearn/linear_model/_sag_fast.pyx.tp']
+    gen_from_templates(templates, top_path)
 
     config.add_extension('_sag_fast',
                          sources=['_sag_fast.pyx'],
diff --git a/sklearn/utils/setup.py b/sklearn/utils/setup.py
index 370bd20e01d69..098adeeccab09 100644
--- a/sklearn/utils/setup.py
+++ b/sklearn/utils/setup.py
@@ -1,11 +1,12 @@
 import os
 from os.path import join
 
+from sklearn._build_utils import gen_from_templates
+
 
 def configuration(parent_package='', top_path=None):
     import numpy
     from numpy.distutils.misc_util import Configuration
-    from Cython import Tempita
 
     config = Configuration('utils', parent_package, top_path)
 
@@ -45,23 +46,10 @@ def configuration(parent_package='', top_path=None):
                          sources=['_openmp_helpers.pyx'],
                          libraries=libraries)
 
-    # generate files from a template
-    pyx_templates = ['sklearn/utils/_seq_dataset.pyx.tp',
-                     'sklearn/utils/_seq_dataset.pxd.tp']
-
-    for pyxfiles in pyx_templates:
-        outfile = pyxfiles.replace('.tp', '')
-        # if .pyx.tp is not updated, no need to output .pyx
-        if (os.path.exists(outfile) and
-                os.stat(pyxfiles).st_mtime < os.stat(outfile).st_mtime):
-            continue
-
-        with open(pyxfiles, "r") as f:
-            tmpl = f.read()
-        pyxcontent = Tempita.sub(tmpl)
-
-        with open(outfile, "w") as f:
-            f.write(pyxcontent)
+    # generate _seq_dataset from template
+    templates = ['sklearn/utils/_seq_dataset.pyx.tp',
+                 'sklearn/utils/_seq_dataset.pxd.tp']
+    gen_from_templates(templates, top_path)
 
     config.add_extension('_seq_dataset',
                          sources=['_seq_dataset.pyx'],