diff --git a/setup.py b/setup.py
index e54316d..2deab95 100644
--- a/setup.py
+++ b/setup.py
@@ -22,6 +22,7 @@
 	],
 	packages = [
 		"sklearn2pmml",
+		"sklearn2pmml.cross_reference",
 		"sklearn2pmml.decoration",
 		"sklearn2pmml.ensemble",
 		"sklearn2pmml.expression",
diff --git a/sklearn2pmml/cross_reference/__init__.py b/sklearn2pmml/cross_reference/__init__.py
new file mode 100644
index 0000000..41f20e3
--- /dev/null
+++ b/sklearn2pmml/cross_reference/__init__.py
@@ -0,0 +1,42 @@
+from sklearn.base import BaseEstimator, TransformerMixin
+
+import numpy
+
+class _BaseMemoryManager(BaseEstimator, TransformerMixin):
+
+	def __init__(self, memory, names):
+		self.memory = memory
+		if not isinstance(names, list):
+			raise TypeError()
+		self.names = names
+
+class Memorizer(_BaseMemoryManager):
+
+	def __init__(self, memory, names):
+		super(Memorizer, self).__init__(memory, names)
+
+	def fit(self, X, y = None):
+		if X.shape[1] != len(self.names):
+			raise ValueError()
+		return self
+
+	def transform(self, X):
+		for idx, name in enumerate(self.names):
+			x = X[:, idx]
+			self.memory[name] = x.copy()
+		return numpy.empty(shape = (X.shape[0], 0), dtype = int)
+
+class Recaller(_BaseMemoryManager):
+
+	def __init__(self, memory, names):
+		super(Recaller, self).__init__(memory, names)
+
+	def fit(self, X, y = None):
+		return self
+
+	def transform(self, X):
+		result = []
+		for idx, name in enumerate(self.names):
+			x = self.memory[name]
+			result.append(x.copy())
+		return numpy.asarray(result).T
diff --git a/sklearn2pmml/cross_reference/tests/__init__.py b/sklearn2pmml/cross_reference/tests/__init__.py
new file mode 100644
index 0000000..2f65453
--- /dev/null
+++ b/sklearn2pmml/cross_reference/tests/__init__.py
@@ -0,0 +1,56 @@
+from pandas import DataFrame
+from sklearn2pmml.cross_reference import Memorizer, Recaller
+from unittest import TestCase
+
+import numpy
+
+class MemorizerTest(TestCase):
+
+	def test_fit_transform(self):
+		memory = dict()
+		self.assertEqual(0, len(memory))
+		memorizer = Memorizer(memory, ["int"])
+		X = numpy.asarray([[-1], [1]])
+		Xt = memorizer.fit_transform(X)
+		self.assertEqual((2, 0), Xt.shape)
+		self.assertEqual(1, len(memory))
+		self.assertEqual([-1, 1], memory["int"].tolist())
+
+		memory = DataFrame()
+		self.assertEqual((0, 0), memory.shape)
+		memorizer = Memorizer(memory, ["int", "float", "str"])
+		X = numpy.asarray([[1, 1.0, "one"], [2, 2.0, "two"], [3, 3.0, "three"]])
+		Xt = memorizer.fit_transform(X)
+		self.assertEqual((3, 0), Xt.shape)
+		self.assertEqual((3, 3), memory.shape)
+		self.assertEqual(["1", "2", "3"], memory["int"].tolist())
+		self.assertEqual([1, 2, 3], memory["int"].astype(int).tolist())
+		self.assertEqual([str(1.0), str(2.0), str(3.0)], memory["float"].tolist())
+		self.assertEqual([1.0, 2.0, 3.0], memory["float"].astype(float).tolist())
+		self.assertEqual(["one", "two", "three"], memory["str"].tolist())
+
+class RecallerTest(TestCase):
+
+	def test_fit_transform(self):
+		X = numpy.empty((100, 5), dtype = str)
+
+		memory = {
+			"int": [-1, 1]
+		}
+		recaller = Recaller(memory, ["int"])
+		Xt = recaller.fit_transform(X)
+		self.assertEqual((2, 1), Xt.shape)
+		self.assertEqual([-1, 1], Xt[:, 0].tolist())
+
+		memory = DataFrame([[1, 1.0, "one"], [2, 2.0, "two"], [3, 3.0, "three"]], columns = ["int", "float", "str"])
+		self.assertEqual((3, 3), memory.shape)
+		recaller = Recaller(memory, ["int"])
+		Xt = recaller.fit_transform(X)
+		self.assertEqual((3, 1), Xt.shape)
+		self.assertEqual([1, 2, 3], Xt[:, 0].tolist())
+		recaller = Recaller(memory, ["int", "float", "str"])
+		Xt = recaller.fit_transform(X)
+		self.assertEqual((3, 3), Xt.shape)
+		self.assertEqual([1, 2, 3], Xt[:, 0].tolist())
+		self.assertEqual([1.0, 2.0, 3.0], Xt[:, 1].tolist())
+		self.assertEqual(["one", "two", "three"], Xt[:, 2].tolist())